def load_net_model(): parser = argparse.ArgumentParser() parser.add_argument('--seg_model', type=str, default='D:/DenseFusion_1.0_20210308/Seg_Pose/Seg/trained_models/model_2_0.0033125885542725133.pth', help='resume Seg model') parser.add_argument('--model', type=str, default='D:/DenseFusion_1.0_20210308/Seg_Pose/Pose/trained_models/pose_model_2_0.09233990000864985.pth', help='resume PoseNet model') parser.add_argument('--refine_model', type=str, default='D:/DenseFusion_1.0_20210308/Seg_Pose/Pose/trained_models/pose_refine_model_4_0.06547849291308346.pth', help='resume PoseRefineNet model') opt = parser.parse_args() global estimator estimator = PoseNet(num_points=num_points, num_obj=num_objects) estimator.cuda() global refiner refiner = PoseRefineNet(num_points=num_points, num_obj=num_objects) refiner.cuda() estimator.load_state_dict(torch.load(opt.model)) refiner.load_state_dict(torch.load(opt.refine_model)) estimator.eval() refiner.eval() global testdataset testdataset = sp_Dataset(opt.seg_model, num_points, False, 0.0, True)
def main(args): seg_model = segnet() seg_model.cuda() #seg_model.load_state_dict(torch.load('./segschaltgabel.pth')) #uncomment if we are using the original segnet #seg_model.load_state_dict(torch.load('./schaltgabel_pruned.pth')) #seg_model.load_state_dict(torch.load('./stift_pruned.pth')) #seg_model.load_state_dict(torch.load('./flansch_pruned.pth')) if opt.model == 'flansch': seg_model.load_state_dict(torch.load('./segflansch.pth')) #seg_model.load_state_dict(torch.load('./flansch_pruned.pth')) idx = 0 scaled = np.array([[59, -42, 59], [59, -42, -59], [59, 20, -59], [59, 20, 59], [-59, -42, 59], [-59, -42, -59], [-59, 20, -59], [-59, 20, 59] ]) / 1000 #flansch_3d_bbox elif opt.model == 'schaltgabel': seg_model.load_state_dict(torch.load('./segschaltgabel.pth')) #seg_model.load_state_dict(torch.load('./schaltgabel_pruned.pth')) idx = 1 scaled = np.array([[-54.7478, -16.7500, 23], [-54.7478, -16.7500, 0], [54.7478, -16.7500, 0], [54.7478, -16.7500, 23], [-54.7478, 130.85, 23], [-54.7478, 130.85, 0], [54.7478, 130.85, 0], [54.7478, 130.85, 23] ]) / 1000 #schaltgabel_3d_bbox else: #seg_model.load_state_dict(torch.load('./stift_pruned.pth')) seg_model.load_state_dict(torch.load('./segstift.pth')) idx = 2 scaled = np.array( [[-20, -35, 19.9878], [-20, -35, -19.9878], [20, -35, -19.9878], [20, -35, 19.9878], [-20, 87, 19.9878], [-20, 87, -19.9878], [20, 87, -19.9878], [20, 87, 19.9878]]) / 1000 #stift_3d_bbox seg_model.eval() estimator = PoseNet(num_points, num_objects) estimator.cuda() refiner = PoseRefineNet(num_points, num_objects) refiner.cuda() estimator.load_state_dict(torch.load('./pose_model_current.pth')) refiner.load_state_dict(torch.load('./pose_refine_model_current.pth')) estimator.eval() refiner.eval() pe = pose_estimation(seg_model, estimator, refiner, idx, scaled) rospy.init_node('pose_estimation', anonymous=True) try: rospy.spin() except KeyboardInterrupt: print('Shutting down ROS pose estimation module') cv2.destroyAllWindows()
def main(): # Globals global refiner, estimator # Setup upload folder and run server if not os.path.exists(UPLOAD_FOLDER): os.makedirs(UPLOAD_FOLDER) # Sets up network refiner = PoseRefineNet(num_points=num_points, num_obj=num_obj) refiner.cuda() refiner.load_state_dict(torch.load(opt.refine_model)) refiner.eval() estimator = PoseNet(num_points=num_points, num_obj=num_obj) estimator.cuda() estimator.load_state_dict(torch.load(opt.model)) estimator.eval() app.config['UPLOAD_FOLDER'] = UPLOAD_FOLDER app.run(port=PORT, host='0.0.0.0', debug=False)
class DenseFusionRefine: def __init__(self, dataset): self.num_points = dataset.df_num_points self.obj_list = dataset.obj_ids self.refiner = PoseRefineNet(num_points=self.num_points, num_obj=len(self.obj_list)) self.refiner.cuda() self.refiner.load_state_dict(torch.load(dataset.df_model_path)) self.refiner.eval() def refine(self, obj_id, pose, emb, cloud, iterations=1): class_id = self.obj_list.index(obj_id) index = torch.LongTensor([class_id]) index = Variable(index).cuda() pose_ref = pose.copy() for _ in range(iterations): # transform cloud according to pose estimate R = Variable(torch.from_numpy(pose_ref[:3, :3].astype( np.float32))).cuda().view(1, 3, 3) ts = Variable(torch.from_numpy(pose_ref[:3, 3].astype(np.float32))).cuda().view(1, 3)\ .repeat(self.num_points, 1).contiguous().view(1, self.num_points, 3) new_cloud = torch.bmm((cloud - ts), R).contiguous() # predict delta pose pred_q, pred_t = self.refiner(new_cloud, emb, index) pred_q = pred_q.view(1, 1, -1) pred_q = pred_q / (torch.norm(pred_q, dim=2).view(1, 1, 1)) pred_q = pred_q.view(-1).cpu().data.numpy() pred_t = pred_t.view(-1).cpu().data.numpy() # apply delta to get new pose estimate pose_delta = quaternion_matrix(pred_q) pose_delta[:3, 3] = pred_t pose_ref = pose_ref @ pose_delta return pose_ref
model_points_list.append(model_points) model = "/home/tomo/densefusion/trained_models/tomo_blisters_2/pose_model_1_0.005430417195209802.pth" refine_model = "/home/tomo/densefusion/trained_models/tomo_blisters_2/pose_refine_model_63_0.0005920357997828468.pth" #model = "/home/tomo/densefusion/trained_models/tomo_blisters_invert/pose_model_1_0.005256925600131684.pth" #refine_model = "/home/tomo/densefusion/trained_models/tomo_blisters_invert/pose_refine_model_29_0.0005998273272780352.pth" estimator = PoseNet(num_points=num_points, num_obj=num_objects) estimator.cuda() refiner = PoseRefineNet(num_points=num_points, num_obj=num_objects) refiner.cuda() estimator.load_state_dict(torch.load(model)) refiner.load_state_dict(torch.load(refine_model)) estimator.eval() refiner.eval() def setup_config(): model_path = "/home/tomo/densefusion/MASK_RCNN_TOMO/trained_models/model_final_new.pth" cfg = get_cfg() cfg.merge_from_file( model_zoo.get_config_file( "COCO-InstanceSegmentation/mask_rcnn_R_101_FPN_3x.yaml")) cfg.INPUT.MIN_SIZE_TEST = 0 cfg.INPUT.MAX_SIZE_TEST = 0 cfg.MODEL.RPN.PRE_NMS_TOPK_TEST = 2000 cfg.MODEL.WEIGHTS = model_path cfg.MODEL.ANCHOR_GENERATOR.SIZES = [[8, 16, 32, 64, 128]] cfg.MODEL.RPN.BATCH_SIZE_PER_IMAGE = 512 cfg.MODEL.ROI_HEADS.BATCH_SIZE_PER_IMAGE = 256
def main(): opt.manualSeed = random.randint(1, 10000) random.seed(opt.manualSeed) torch.manual_seed(opt.manualSeed) if opt.dataset == 'ycb': opt.num_objects = 21 #number of object classes in the dataset opt.num_points = 1000 #number of points on the input pointcloud opt.outf = 'trained_models/ycb' #folder to save trained models opt.log_dir = 'experiments/logs/ycb' #folder to save logs opt.repeat_epoch = 1 #number of repeat times for one epoch training elif opt.dataset == 'linemod': opt.num_objects = 13 opt.num_points = 500 opt.outf = 'trained_models/linemod' opt.log_dir = 'experiments/logs/linemod' opt.repeat_epoch = 20 else: print('Unknown dataset') return estimator = PoseNet(num_points=opt.num_points, num_obj=opt.num_objects) estimator.cuda() refiner = PoseRefineNet(num_points=opt.num_points, num_obj=opt.num_objects) refiner.cuda() if opt.resume_posenet != '': estimator.load_state_dict( torch.load('{0}/{1}'.format(opt.outf, opt.resume_posenet))) if opt.resume_refinenet != '': refiner.load_state_dict( torch.load('{0}/{1}'.format(opt.outf, opt.resume_refinenet))) opt.refine_start = True opt.decay_start = True opt.lr *= opt.lr_rate opt.w *= opt.w_rate opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) else: opt.refine_start = False opt.decay_start = False optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) if opt.dataset == 'ycb': dataset = PoseDataset_ycb('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'linemod': dataset = PoseDataset_linemod('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) if opt.dataset == 'ycb': test_dataset = PoseDataset_ycb('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'linemod': test_dataset = PoseDataset_linemod('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) opt.sym_list = dataset.get_sym_list() opt.num_points_mesh = dataset.get_num_points_mesh() print( '>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}\nsymmetry object list: {3}' .format(len(dataset), len(test_dataset), opt.num_points_mesh, opt.sym_list)) criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list) best_test = np.Inf if opt.start_epoch == 1: for log in os.listdir(opt.log_dir): os.remove(os.path.join(opt.log_dir, log)) st_time = time.time() for epoch in range(opt.start_epoch, opt.nepoch): logger = setup_logger( 'epoch%d' % epoch, os.path.join(opt.log_dir, 'epoch_%d_log.txt' % epoch)) logger.info('Train time {0}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Training started')) train_count = 0 train_dis_avg = 0.0 if opt.refine_start: estimator.eval() refiner.train() else: estimator.train() optimizer.zero_grad() for rep in range(opt.repeat_epoch): for i, data in enumerate(dataloader, 0): points, choose, img, target, model_points, idx = data points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator( img, points, choose, idx) loss, dis, new_points, new_target = criterion( pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine( pred_r, pred_t, new_target, model_points, idx, new_points) dis.backward() else: loss.backward() train_dis_avg += dis.item() train_count += 1 if train_count % opt.batch_size == 0: logger.info( 'Train time {0} Epoch {1} Batch {2} Frame {3} Avg_dis:{4}' .format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, int(train_count / opt.batch_size), train_count, train_dis_avg / opt.batch_size)) optimizer.step() optimizer.zero_grad() train_dis_avg = 0 if train_count != 0 and train_count % 1000 == 0: if opt.refine_start: torch.save( refiner.state_dict(), '{0}/pose_refine_model_current.pth'.format( opt.outf)) else: torch.save( estimator.state_dict(), '{0}/pose_model_current.pth'.format(opt.outf)) print( '>>>>>>>>----------epoch {0} train finish---------<<<<<<<<'.format( epoch)) logger = setup_logger( 'epoch%d_test' % epoch, os.path.join(opt.log_dir, 'epoch_%d_test_log.txt' % epoch)) logger.info('Test time {0}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Testing started')) test_dis = 0.0 test_count = 0 estimator.eval() refiner.eval() for j, data in enumerate(testdataloader, 0): points, choose, img, target, model_points, idx = data points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator(img, points, choose, idx) _, dis, new_points, new_target = criterion(pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine( pred_r, pred_t, new_target, model_points, idx, new_points) test_dis += dis.item() logger.info('Test time {0} Test Frame No.{1} dis:{2}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), test_count, dis)) test_count += 1 test_dis = test_dis / test_count logger.info('Test time {0} Epoch {1} TEST FINISH Avg dis: {2}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, test_dis)) if test_dis <= best_test: best_test = test_dis if opt.refine_start: torch.save( refiner.state_dict(), '{0}/pose_refine_model_{1}_{2}.pth'.format( opt.outf, epoch, test_dis)) else: torch.save( estimator.state_dict(), '{0}/pose_model_{1}_{2}.pth'.format( opt.outf, epoch, test_dis)) print(epoch, '>>>>>>>>----------BEST TEST MODEL SAVED---------<<<<<<<<') if best_test < opt.decay_margin and not opt.decay_start: opt.decay_start = True opt.lr *= opt.lr_rate opt.w *= opt.w_rate optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) if best_test < opt.refine_margin and not opt.refine_start: opt.refine_start = True opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) if opt.dataset == 'ycb': dataset = PoseDataset_ycb('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'linemod': dataset = PoseDataset_linemod('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) if opt.dataset == 'ycb': test_dataset = PoseDataset_ycb('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'linemod': test_dataset = PoseDataset_linemod('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) testdataloader = torch.utils.data.DataLoader( test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) opt.sym_list = dataset.get_sym_list() opt.num_points_mesh = dataset.get_num_points_mesh() print( '>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}\nsymmetry object list: {3}' .format(len(dataset), len(test_dataset), opt.num_points_mesh, opt.sym_list)) criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list)
class DenseFusionEstimator(object): def __init__(self, num_points, num_obj, estimator_weights_file, refiner_weights_file=None, batch_size=1): self.num_points = num_points self.bs = batch_size self.estimator = PoseNet(num_points=num_points, num_obj=num_obj) if (torch.cuda.is_available()): self.estimator.cuda() self.estimator.load_state_dict(torch.load(estimator_weights_file)) else: self.estimator.load_state_dict( torch.load(estimator_weights_file, map_location=lambda storage, loc: storage)) self.estimator.eval() if (refiner_weights_file is not None): self.refiner = PoseRefineNet(num_points=num_points, num_obj=num_obj) if (torch.cuda.is_available()): self.refiner.cuda() self.refiner.load_state_dict(torch.load(refiner_weights_file)) self.refiner.eval() else: self.refiner = None def __call__(self, img, depth, mask, bbox, object_label, return_all=False): object_label = Variable(torch.LongTensor([object_label])) if (torch.cuda.is_available()): object_label = object_label.cuda() data = preprocessData(img, depth, mask, bbox, num_points=self.num_points) if (data is None): return None, None img_masked, cloud, choose = data max_r, max_t, max_c, pred_r, pred_t, pred_c, emb = self.estimatePose( img_masked, cloud, choose, object_label, return_all=True) if (self.refiner is not None): refined_r, refined_t = self.refinePose(emb, cloud, object_label, max_r, max_t) return refined_r, refined_t if (return_all): return max_r, max_t, max_c, pred_r, pred_t, pred_c, emb return max_r, max_t def globalFeature(self, img_masked, cloud, choose, object_label): feat = self.estimator.globalFeature(img_masked, cloud, choose, object_label) return feat def estimatePose(self, img_masked, cloud, choose, object_label, return_all=True): pred_r, pred_t, pred_c, emb = self.estimator(img_masked, cloud, choose, object_label) pred_r = pred_r / torch.norm(pred_r, dim=2).view(1, self.num_points, 1) pred_c = pred_c.view(self.bs, self.num_points) max_c, which_max = torch.max(pred_c, 1) pred_t = pred_t.view(self.bs * self.num_points, 1, 3) points = cloud.view(self.bs * self.num_points, 1, 3) max_r = pred_r[0][which_max[0]].view(-1) #.cpu().data.numpy() max_t = (points + pred_t)[which_max[0]].view(-1) #.cpu().data.numpy() #max_c = max_c.cpu().data.numpy() if (return_all): return max_r, max_t, max_c, pred_r, pred_t, pred_c, emb return max_r, max_t, max_c def refinePose(self, emb, cloud, object_label, init_t, init_r, iterations=2): init_t = init_t.cpu().data.numpy() init_r = init_r.cpu().data.numpy() for ite in range(0, iteration): T = Variable(torch.from_numpy( init_t.astype(np.float32))).cuda().view(1, 3).repeat( num_points, 1).contiguous().view(1, self.num_points, 3) init_mat = quaternion_matrix(init_r) R = Variable(torch.from_numpy(init_mat[:3, :3].astype( np.float32))).cuda().view(1, 3, 3) init_mat[0:3, 3] = init_t new_cloud = torch.bmm((cloud - T), R).contiguous() pred_r, pred_t = self.refiner(new_cloud, emb, object_label) pred_r = pred_r.view(1, 1, -1) pred_r = pred_r / (torch.norm(pred_r, dim=2).view(1, 1, 1)) delta_r = pred_r.view(-1).cpu().data.numpy() delta_t = pred_t.view(-1).cpu().data.numpy() delta_mat = quaternion_matrix(delta_r) delta_mat[0:3, 3] = delta_t refined_mat = np.dot(init_mat, delta_mat) refined_r = copy.deepcopy(refined_mat) refined_r[0:3, 3] = 0 refined_r = quaternion_from_matrix(refined_r, True) refined_t = np.array( [refined_mat[0][3], refined_mat[1][3], refined_mat[2][3]]) init_r = r_final init_t = t_final return refined_t, refined_t
class Inference: def __init__(self, model, model_with_refinement, width, height, fx, fy, cx, cy, number_points=1000, number_iterations=2): self.width = width self.height = height self.fx = fx self.fy = fy self.cx = cx self.cy = cy self.number_iterations = number_iterations self.norm = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) self.border_list = [ -1, 40, 80, 120, 160, 200, 240, 280, 320, 360, 400, 440, 480, 520, 560, 600, 640, 680 ] self.xmap = numpy.array([[j for i in range(self.width)] for j in range(self.height)]) self.ymap = numpy.array([[i for i in range(self.width)] for j in range(self.height)]) self.number_points = number_points self.estimator = PoseNet(num_points=self.number_points, num_obj=21) self.estimator.cuda() self.estimator.load_state_dict(torch.load(model)) self.estimator.eval() self.refiner = PoseRefineNet(num_points=self.number_points, num_obj=21) self.refiner.cuda() self.refiner.load_state_dict(torch.load(model_with_refinement)) self.refiner.eval() def evaluate_bbox(self, mask): points = cv2.findNonZero(mask) x, y, w, h = cv2.boundingRect(points) rmin = y rmax = y + h cmin = x cmax = x + w r_b = rmax - rmin for tt in range(len(self.border_list)): if r_b > self.border_list[tt] and r_b < self.border_list[tt + 1]: r_b = self.border_list[tt + 1] break c_b = cmax - cmin for tt in range(len(self.border_list)): if c_b > self.border_list[tt] and c_b < self.border_list[tt + 1]: c_b = self.border_list[tt + 1] break center = [int((rmin + rmax) / 2), int((cmin + cmax) / 2)] rmin = center[0] - int(r_b / 2) rmax = center[0] + int(r_b / 2) cmin = center[1] - int(c_b / 2) cmax = center[1] + int(c_b / 2) if rmin < 0: delt = -rmin rmin = 0 rmax += delt if cmin < 0: delt = -cmin cmin = 0 cmax += delt if rmax > self.height: delt = rmax - self.height rmax = self.height rmin -= delt if cmax > self.width: delt = cmax - self.width cmax = self.width cmin -= delt return rmin, rmax, cmin, cmax def inference(self, object_id, img, depth, mask): try: rmin, rmax, cmin, cmax = self.evaluate_bbox(mask) mask_depth = numpy_ma.getmaskarray( numpy_ma.masked_not_equal(depth, 0)) mask_label = (mask == 255) mask = mask_label * mask_depth choose = mask[rmin:rmax, cmin:cmax].flatten().nonzero()[0] if len(choose) > self.number_points: c_mask = numpy.zeros(len(choose), dtype=int) c_mask[:self.number_points] = 1 numpy.random.shuffle(c_mask) choose = choose[c_mask.nonzero()] else: choose = numpy.pad(choose, (0, self.number_points - len(choose)), 'wrap') depth_masked = depth[ rmin:rmax, cmin:cmax].flatten()[choose][:, numpy.newaxis].astype( numpy.float32) xmap_masked = self.xmap[ rmin:rmax, cmin:cmax].flatten()[choose][:, numpy.newaxis].astype( numpy.float32) ymap_masked = self.ymap[ rmin:rmax, cmin:cmax].flatten()[choose][:, numpy.newaxis].astype( numpy.float32) choose = numpy.array([choose]) pt2 = depth_masked pt0 = (ymap_masked - self.cx) * pt2 / self.fx pt1 = (xmap_masked - self.cy) * pt2 / self.fy cloud = numpy.concatenate((pt0, pt1, pt2), axis=1) img_masked = numpy.array(img)[:, :, :3] img_masked = numpy.transpose(img_masked, (2, 0, 1)) img_masked = img_masked[:, rmin:rmax, cmin:cmax] cloud = torch.from_numpy(cloud.astype(numpy.float32)) choose = torch.LongTensor(choose.astype(numpy.int32)) img_masked = self.norm( torch.from_numpy(img_masked.astype(numpy.float32))) index = torch.LongTensor([object_id - 1]) cloud = Variable(cloud).cuda() choose = Variable(choose).cuda() img_masked = Variable(img_masked).cuda() index = Variable(index).cuda() cloud = cloud.view(1, self.number_points, 3) img_masked = img_masked.view(1, 3, img_masked.size()[1], img_masked.size()[2]) pred_r, pred_t, pred_c, emb = self.estimator( img_masked, cloud, choose, index) pred_r = pred_r / torch.norm(pred_r, dim=2).view( 1, self.number_points, 1) pred_c = pred_c.view(1, self.number_points) how_max, which_max = torch.max(pred_c, 1) pred_t = pred_t.view(1 * self.number_points, 1, 3) points = cloud.view(1 * self.number_points, 1, 3) my_r = pred_r[0][which_max[0]].view(-1).cpu().data.numpy() my_t = (points + pred_t)[which_max[0]].view(-1).cpu().data.numpy() my_pred = numpy.append(my_r, my_t) for ite in range(0, self.number_iterations): T = Variable(torch.from_numpy(my_t.astype( numpy.float32))).cuda().view(1, 3).repeat( self.number_points, 1).contiguous().view(1, self.number_points, 3) my_mat = quaternion_matrix(my_r) R = Variable( torch.from_numpy(my_mat[:3, :3].astype( numpy.float32))).cuda().view(1, 3, 3) my_mat[0:3, 3] = my_t new_cloud = torch.bmm((cloud - T), R).contiguous() pred_r, pred_t = self.refiner(new_cloud, emb, index) pred_r = pred_r.view(1, 1, -1) pred_r = pred_r / (torch.norm(pred_r, dim=2).view(1, 1, 1)) my_r_2 = pred_r.view(-1).cpu().data.numpy() my_t_2 = pred_t.view(-1).cpu().data.numpy() my_mat_2 = quaternion_matrix(my_r_2) my_mat_2[0:3, 3] = my_t_2 my_mat_final = numpy.dot(my_mat, my_mat_2) my_r_final = copy.deepcopy(my_mat_final) my_r_final[0:3, 3] = 0 my_r_final_copy = copy.deepcopy(my_r_final) my_r_final = quaternion_from_matrix(my_r_final, True) my_t_final = numpy.array([ my_mat_final[0][3], my_mat_final[1][3], my_mat_final[2][3] ]) my_pred = numpy.append(my_r_final, my_t_final) my_r_final_alt = Quaternion(matrix=my_r_final_copy[0:3, 0:3]) my_r_final_alt_aa = numpy.append(my_r_final_alt.axis.squeeze(), my_r_final_alt.angle) my_pred_alt = numpy.append(my_t_final, my_r_final_alt_aa) my_r = my_r_final my_t = my_t_final return my_pred_alt except ValueError: print( "Inference::inference. Error: an error occured at inference time." ) return []
def main(): opt.manualSeed = random.randint(1, 10000) random.seed(opt.manualSeed) torch.manual_seed(opt.manualSeed) opt.num_objects = 21 #number of object classes in the dataset opt.num_points = 1000 #number of points on the input pointcloud opt.outf = 'trained_models/ycb_global_mnorm' #folder to save trained models opt.log_dir = 'experiments/logs/ycb_global_mnorm' #folder to save logs opt.repeat_epoch = 1 #number of repeat times for one epoch training if not os.path.exists(opt.outf): os.makedirs(opt.outf) if not os.path.exists(opt.log_dir): os.makedirs(opt.log_dir) estimator = PoseNetGlobal(num_points = opt.num_points, num_obj = opt.num_objects) estimator.cuda() refiner = PoseRefineNet(num_points = opt.num_points, num_obj = opt.num_objects) refiner.cuda() if opt.resume_posenet != '': estimator.load_state_dict(torch.load('{0}/{1}'.format(opt.outf, opt.resume_posenet))) if opt.resume_refinenet != '': refiner.load_state_dict(torch.load('{0}/{1}'.format(opt.outf, opt.resume_refinenet))) opt.refine_start = True opt.decay_start = True opt.lr *= opt.lr_rate opt.w *= opt.w_rate opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) else: opt.refine_start = False opt.decay_start = False optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) object_list = list(range(1,22)) output_format = [otypes.DEPTH_POINTS_MASKED_AND_INDEXES, otypes.IMAGE_CROPPED, otypes.MODEL_POINTS_TRANSFORMED, otypes.MODEL_POINTS, otypes.OBJECT_LABEL, ] dataset = YCBDataset(opt.dataset_root, mode='train_syn_grid', object_list = object_list, output_data = output_format, resample_on_error = True, preprocessors = [YCBOcclusionAugmentor(opt.dataset_root), ColorJitter(), InplaneRotator()], postprocessors = [ImageNormalizer(), PointMeanNormalizer(0)], #PointShifter()], refine = opt.refine_start, image_size = [640, 480], num_points=1000) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers-1) test_dataset = YCBDataset(opt.dataset_root, mode='valid', object_list = object_list, output_data = output_format, resample_on_error = True, preprocessors = [], postprocessors = [ImageNormalizer(), PointMeanNormalizer(0)], refine = opt.refine_start, image_size = [640, 480], num_points=1000) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=1) opt.sym_list = [12, 15, 18, 19, 20] opt.num_points_mesh = dataset.num_pt_mesh_small print('>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}\nsymmetry object list: {3}'.format(len(dataset), len(test_dataset), opt.num_points_mesh, opt.sym_list)) criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list) best_test = np.Inf if opt.start_epoch == 1: for log in os.listdir(opt.log_dir): os.remove(os.path.join(opt.log_dir, log)) st_time = time.time() for epoch in range(opt.start_epoch, opt.nepoch): logger = setup_logger('epoch%d' % epoch, os.path.join(opt.log_dir, 'epoch_%d_log.txt' % epoch)) logger.info('Train time {0}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Training started')) train_count = 0 train_dis_avg = 0.0 if opt.refine_start: estimator.eval() refiner.train() else: estimator.train() optimizer.zero_grad() for rep in range(opt.repeat_epoch): for i, data in enumerate(dataloader, 0): points, choose, img, target, model_points, idx = data idx = idx - 1 points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator(img, points, choose, idx) loss, dis, new_points, new_target = criterion(pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_target, model_points, idx, new_points) dis.backward() else: loss.backward() train_dis_avg += dis.item() train_count += 1 if train_count % opt.batch_size == 0: logger.info('Train time {0} Epoch {1} Batch {2} Frame {3} Avg_dis:{4}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, int(train_count / opt.batch_size), train_count, train_dis_avg / opt.batch_size)) optimizer.step() optimizer.zero_grad() train_dis_avg = 0 if train_count != 0 and train_count % 1000 == 0: if opt.refine_start: torch.save(refiner.state_dict(), '{0}/pose_refine_model_current.pth'.format(opt.outf)) else: torch.save(estimator.state_dict(), '{0}/pose_model_current.pth'.format(opt.outf)) print('>>>>>>>>----------epoch {0} train finish---------<<<<<<<<'.format(epoch)) logger = setup_logger('epoch%d_test' % epoch, os.path.join(opt.log_dir, 'epoch_%d_test_log.txt' % epoch)) logger.info('Test time {0}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Testing started')) test_dis = 0.0 test_count = 0 estimator.eval() refiner.eval() for j, data in enumerate(testdataloader, 0): points, choose, img, target, model_points, idx = data idx = idx - 1 points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator(img, points, choose, idx) _, dis, new_points, new_target = criterion(pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_target, model_points, idx, new_points) test_dis += dis.item() logger.info('Test time {0} Test Frame No.{1} dis:{2}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), test_count, dis)) test_count += 1 test_dis = test_dis / test_count logger.info('Test time {0} Epoch {1} TEST FINISH Avg dis: {2}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, test_dis)) if test_dis <= best_test: best_test = test_dis if opt.refine_start: torch.save(refiner.state_dict(), '{0}/pose_refine_model_{1}_{2}.pth'.format(opt.outf, epoch, test_dis)) else: torch.save(estimator.state_dict(), '{0}/pose_model_{1}_{2}.pth'.format(opt.outf, epoch, test_dis)) print(epoch, '>>>>>>>>----------BEST TEST MODEL SAVED---------<<<<<<<<') if best_test < opt.decay_margin and not opt.decay_start: opt.decay_start = True opt.lr *= opt.lr_rate opt.w *= opt.w_rate optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) if best_test < opt.refine_margin and not opt.refine_start: opt.refine_start = True opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) dataset = YCBDataset(opt.dataset_root, mode='train_syn_grid', object_list = object_list, output_data = output_format, resample_on_error = True, preprocessors = [YCBOcclusionAugmentor(opt.dataset_root), ColorJitter(), InplaneRotator()], postprocessors = [ImageNormalizer(), PointShifter()], refine = opt.refine_start, image_size = [640, 480], num_points=1000) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) test_dataset = YCBDataset(opt.dataset_root, mode='valid', object_list = object_list, output_data = output_format, resample_on_error = True, preprocessors = [], postprocessors = [ImageNormalizer()], refine = opt.refine_start, image_size = [640, 480], num_points=1000) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) opt.num_points_mesh = dataset.num_pt_mesh_large print('>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}\nsymmetry object list: {3}'.format(len(dataset), len(test_dataset), opt.num_points_mesh, opt.sym_list)) criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list)
class DenseFusion_YW(object): def __init__(self, posecnn): # posecnn self.posecnn = posecnn # df #camera paras self.cam_cx = 312.9869 #??? self.cam_cy = 241.3109 #??? self.cam_fx = 1066.778 #x方向焦距,水平 self.cam_fy = 1067.487 #y方向焦距,垂直 self.cam_scale = 10000.0 # scale ???? #main self.num_points = 1000 # default self.num_obj = 21 # ycb 已经训练好了,含有21个类,不能改。 self.xmap = np.array([[j for i in range(640)] for j in range(480) ]) # [0000...,1111...,222] 480,640 self.ymap = np.array([ [i for i in range(640)] for j in range(480) ]) # [0 1 2 3..., 0 1 2 3 ..., 0 1 2 3 ...] 480,640 self.norm = transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225 ]) # For img transform in the df self.bs = 1 self.iteration = 2 self.esti_model_file = "src/pose_model_26_0.012863246640872631.pth" self.refine_model_file = "src/pose_refine_model_69_0.009449292959118935.pth" self.estimator = PoseNet(num_points=self.num_points, num_obj=self.num_obj) self.estimator.cuda() self.estimator.load_state_dict(torch.load(self.esti_model_file)) self.estimator.eval() self.refiner = PoseRefineNet(num_points=self.num_points, num_obj=self.num_obj) self.refiner.cuda() self.refiner.load_state_dict(torch.load(self.refine_model_file)) self.refiner.eval() self.lst = [ 1 ] # assume only banana. This is designed for multi-label task, if multi label detected in one img, #they may allocated with different id, id==1.it means the label is 1, id==2, it means the label is 2; so with known the #label, we can know label1 is banana ,label2 is apple like this. self.my_result_wo_refine = [] # ??? self.my_result = [] def DenseFusion(self, img, depth, posecnn_res): my_result_wo_refine = [] itemid = 1 # this is simplified for single label decttion, if multi-label used, check DFYW3.py for more depth = np.array(depth) # img = img seg_res = posecnn_res x1, y1, x2, y2 = seg_res["box"] banana_bbox_draw = self.posecnn.get_box_rcwh(seg_res["box"]) rmin, rmax, cmin, cmax = int(x1), int(x2), int(y1), int(y2) depth = depth[:, :, 1] # because depth has 3 dimensions RGB but they are the all the same with each other mask_depth = ma.getmaskarray(ma.masked_not_equal(depth, 0)) # ok label_banana = np.squeeze(seg_res["mask"]) label_banana = ma.getmaskarray(ma.masked_greater(label_banana, 0.5)) label_banana_nonzeros = label_banana.flatten().nonzero() mask_label = ma.getmaskarray(ma.masked_equal( label_banana, itemid)) # label from banana label mask = mask_label * mask_depth mask_nonzeros = mask[:].flatten().nonzero() choose = mask[rmin:rmax, cmin:cmax].flatten().nonzero()[0] if len(choose) > self.num_points: c_mask = np.zeros(len(choose), dtype=int) c_mask[:self.num_points] = 1 np.random.shuffle(c_mask) choose = choose[c_mask.nonzero()] else: print("len of choose is 0, check error") choose = np.pad(choose, (0, self.num_points - len(choose)), 'wrap') depth_masked = depth[rmin:rmax, cmin:cmax].flatten()[choose][:, np.newaxis].astype( np.float32) xmap_masked = self.xmap[ rmin:rmax, cmin:cmax].flatten()[choose][:, np.newaxis].astype(np.float32) ymap_masked = self.ymap[ rmin:rmax, cmin:cmax].flatten()[choose][:, np.newaxis].astype(np.float32) choose = np.array([choose]) pt2 = depth_masked / self.cam_scale pt0 = (ymap_masked - self.cam_cx) * pt2 / self.cam_fx pt1 = (xmap_masked - self.cam_cy) * pt2 / self.cam_fy cloud = np.concatenate((pt0, pt1, pt2), axis=1) img_np = np.array(img) img_masked = np.array(img)[:, :, :3] img_masked = np.transpose(img_masked, (2, 0, 1)) img_masked = img_masked[:, rmin:rmax, cmin:cmax] cloud = torch.from_numpy(cloud.astype(np.float32)) choose = torch.LongTensor(choose.astype(np.int32)) img_masked = self.norm(torch.from_numpy(img_masked.astype(np.float32))) index = torch.LongTensor([itemid - 1]) cloud = Variable(cloud).cuda() choose = Variable(choose).cuda() img_masked = Variable(img_masked).cuda() index = Variable(index).cuda() cloud = cloud.view(1, self.num_points, 3) img_masked = img_masked.view(1, 3, img_masked.size()[1], img_masked.size()[2]) pred_r, pred_t, pred_c, emb = self.estimator(img_masked, cloud, choose, index) pred_r = pred_r / torch.norm(pred_r, dim=2).view(1, self.num_points, 1) pred_c = pred_c.view(self.bs, self.num_points) how_max, which_max = torch.max(pred_c, 1) pred_t = pred_t.view(self.bs * self.num_points, 1, 3) points = cloud.view(self.bs * self.num_points, 1, 3) my_r = pred_r[0][which_max[0]].view(-1).cpu().data.numpy() my_t = (points + pred_t)[which_max[0]].view(-1).cpu().data.numpy() my_pred = np.append(my_r, my_t) my_result_wo_refine.append(my_pred.tolist()) my_result = [] for ite in range(0, self.iteration): T = Variable(torch.from_numpy( my_t.astype(np.float32))).cuda().view(1, 3).repeat( self.num_points, 1).contiguous().view(1, self.num_points, 3) my_mat = quaternion_matrix(my_r) R = Variable(torch.from_numpy(my_mat[:3, :3].astype( np.float32))).cuda().view(1, 3, 3) my_mat[0:3, 3] = my_t new_cloud = torch.bmm((cloud - T), R).contiguous() pred_r, pred_t = self.refiner(new_cloud, emb, index) pred_r = pred_r.view(1, 1, -1) pred_r = pred_r / (torch.norm(pred_r, dim=2).view(1, 1, 1)) my_r_2 = pred_r.view(-1).cpu().data.numpy() my_t_2 = pred_t.view(-1).cpu().data.numpy() my_mat_2 = quaternion_matrix(my_r_2) my_mat_2[0:3, 3] = my_t_2 my_mat_final = np.dot(my_mat, my_mat_2) my_r_final = copy.deepcopy(my_mat_final) my_r_final[0:3, 3] = 0 my_r_final = quaternion_from_matrix(my_r_final, True) my_t_final = np.array( [my_mat_final[0][3], my_mat_final[1][3], my_mat_final[2][3]]) my_pred = np.append(my_r_final, my_t_final) my_result.append(my_pred.tolist()) my_result_np = np.array(my_result) my_result_mean = np.mean(my_result, axis=0) my_r = my_result_mean[:4] my_t = my_result_mean[4:] my_r_quaternion = my_r return my_r_quaternion, my_t
def main(): # g13: parameter setting ------------------- batch_id = 1 opt.dataset ='linemod' opt.dataset_root = './datasets/linemod/Linemod_preprocessed' estimator_path = 'trained_checkpoints/linemod/pose_model_9_0.01310166542980859.pth' refiner_path = 'trained_checkpoints/linemod/pose_refine_model_493_0.006761023565178073.pth' opt.resume_posenet = estimator_path opt.resume_posenet = refiner_path dataset_config_dir = 'datasets/linemod/dataset_config' output_result_dir = 'experiments/eval_result/linemod' bs = 1 #fixed because of the default setting in torch.utils.data.DataLoader opt.iteration = 2 #default is 4 in eval_linemod.py t1_idx = 0 t1_total_eval_num = 3 axis_range = 0.1 # the length of X, Y, and Z axis in 3D vimg_dir = 'verify_img' if not os.path.exists(vimg_dir): os.makedirs(vimg_dir) #------------------------------------------- if opt.dataset == 'ycb': opt.num_objects = 21 #number of object classes in the dataset opt.num_points = 1000 #number of points on the input pointcloud opt.outf = 'trained_models/ycb' #folder to save trained models opt.log_dir = 'experiments/logs/ycb' #folder to save logs opt.repeat_epoch = 1 #number of repeat times for one epoch training elif opt.dataset == 'linemod': opt.num_objects = 13 opt.num_points = 500 opt.outf = 'trained_models/linemod' opt.log_dir = 'experiments/logs/linemod' opt.repeat_epoch = 20 else: print('Unknown dataset') return estimator = PoseNet(num_points = opt.num_points, num_obj = opt.num_objects) estimator.cuda() refiner = PoseRefineNet(num_points = opt.num_points, num_obj = opt.num_objects) refiner.cuda() if opt.resume_posenet != '': estimator.load_state_dict(torch.load(estimator_path)) if opt.resume_refinenet != '': refiner.load_state_dict(torch.load(refiner_path)) opt.refine_start = True opt.decay_start = True opt.lr *= opt.lr_rate opt.w *= opt.w_rate opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) else: opt.refine_start = False opt.decay_start = False optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) if opt.dataset == 'ycb': test_dataset = PoseDataset_ycb('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'linemod': test_dataset = PoseDataset_linemod('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) print('complete loading testing loader\n') opt.sym_list = test_dataset.get_sym_list() opt.num_points_mesh = test_dataset.get_num_points_mesh() print('>>>>>>>>----------Dataset loaded!---------<<<<<<<<\n\ length of the testing set: {0}\nnumber of sample points on mesh: {1}\n\ symmetry object list: {2}'\ .format( len(test_dataset), opt.num_points_mesh, opt.sym_list)) #load pytorch model estimator.eval() refiner.eval() criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list) fw = open('{0}/t1_eval_result_logs.txt'.format(output_result_dir), 'w') #Pose estimation for j, data in enumerate(testdataloader, 0): # g13: modify this part for evaluation target-------------------- if j == t1_total_eval_num: break #---------------------------------------------------------------- points, choose, img, target, model_points, idx = data if len(points.size()) == 2: print('No.{0} NOT Pass! Lost detection!'.format(j)) fw.write('No.{0} NOT Pass! Lost detection!\n'.format(j)) continue points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator(img, points, choose, idx) _, dis, new_points, new_target = criterion(pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) #if opt.refine_start: #iterative poserefinement # for ite in range(0, opt.iteration): # pred_r, pred_t = refiner(new_points, emb, idx) # dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_target, model_points, idx, new_points) pred_r = pred_r / torch.norm(pred_r, dim=2).view(1, opt.num_points, 1) pred_c = pred_c.view(bs, opt.num_points) how_max, which_max = torch.max(pred_c, 1) pred_t = pred_t.view(bs * opt.num_points, 1, 3) my_r = pred_r[0][which_max[0]].view(-1).cpu().data.numpy() my_t = (points.view(bs * opt.num_points, 1, 3) + pred_t)[which_max[0]].view(-1).cpu().data.numpy() my_pred = np.append(my_r, my_t) for ite in range(0, opt.iteration): T = Variable(torch.from_numpy(my_t.astype(np.float32))).cuda().view(1, 3).repeat(opt.num_points, 1).contiguous().view(1, opt.num_points, 3) my_mat = quaternion_matrix(my_r) R = Variable(torch.from_numpy(my_mat[:3, :3].astype(np.float32))).cuda().view(1, 3, 3) my_mat[0:3, 3] = my_t new_points = torch.bmm((points - T), R).contiguous() pred_r, pred_t = refiner(new_points, emb, idx) pred_r = pred_r.view(1, 1, -1) pred_r = pred_r / (torch.norm(pred_r, dim=2).view(1, 1, 1)) my_r_2 = pred_r.view(-1).cpu().data.numpy() my_t_2 = pred_t.view(-1).cpu().data.numpy() my_mat_2 = quaternion_matrix(my_r_2) my_mat_2[0:3, 3] = my_t_2 my_mat_final = np.dot(my_mat, my_mat_2) my_r_final = copy.deepcopy(my_mat_final) my_r_final[0:3, 3] = 0 my_r_final = quaternion_from_matrix(my_r_final, True) my_t_final = np.array([my_mat_final[0][3], my_mat_final[1][3], my_mat_final[2][3]]) my_pred = np.append(my_r_final, my_t_final) my_r = my_r_final my_t = my_t_final # g13: start drawing pose on image------------------------------------ # pick up image print("index {0}: {1}".format(j, test_dataset.list_rgb[j])) img = Image.open(test_dataset.list_rgb[j]) # pick up center position by bbox meta_file = open('{0}/data/{1}/gt.yml'.format(opt.dataset_root, '%02d' % test_dataset.list_obj[j]), 'r') meta = {} meta = yaml.load(meta_file) which_item = test_dataset.list_rank[j] bbx = meta[which_item][0]['obj_bb'] draw = ImageDraw.Draw(img) # draw box (ensure this is the right object) draw.line((bbx[0],bbx[1], bbx[0], bbx[1]+bbx[3]), fill=(255,0,0), width=5) draw.line((bbx[0],bbx[1], bbx[0]+bbx[2], bbx[1]), fill=(255,0,0), width=5) draw.line((bbx[0],bbx[1]+bbx[3], bbx[0]+bbx[2], bbx[1]+bbx[3]), fill=(255,0,0), width=5) draw.line((bbx[0]+bbx[2],bbx[1], bbx[0]+bbx[2], bbx[1]+bbx[3]), fill=(255,0,0), width=5) #get center c_x = bbx[0]+int(bbx[2]/2) c_y = bbx[1]+int(bbx[3]/2) draw.point((c_x,c_y), fill=(255,255,0)) #get the 3D position of center cam_intrinsic = np.zeros((3,3)) cam_intrinsic.itemset(0, test_dataset.cam_fx) cam_intrinsic.itemset(4, test_dataset.cam_fy) cam_intrinsic.itemset(2, test_dataset.cam_cx) cam_intrinsic.itemset(5, test_dataset.cam_cy) cam_intrinsic.itemset(8, 1) cam_extrinsic = my_mat_final[0:3, :] cam2d_3d = np.matmul(cam_intrinsic, cam_extrinsic) cen_3d = np.matmul(np.linalg.pinv(cam2d_3d), [[c_x],[c_y],[1]]) # replace img.show() with plt.imshow(img) #transpose three 3D axis point into 2D x_3d = cen_3d + [[axis_range],[0],[0],[0]] y_3d = cen_3d + [[0],[axis_range],[0],[0]] z_3d = cen_3d + [[0],[0],[axis_range],[0]] x_2d = np.matmul(cam2d_3d, x_3d) y_2d = np.matmul(cam2d_3d, y_3d) z_2d = np.matmul(cam2d_3d, z_3d) #draw the axis on 2D draw.line((c_x, c_y, x_2d[0], x_2d[1]), fill=(255,255,0), width=5) draw.line((c_x, c_y, y_2d[0], y_2d[1]), fill=(0,255,0), width=5) draw.line((c_x, c_y, z_2d[0], z_2d[1]), fill=(0,0,255), width=5) #g13: show image #img.show() #save file under file img_file_name = '{0}/pred_obj{1}_pic{2}.png'.format(vimg_dir, test_dataset.list_obj[j], which_item) img.save( img_file_name, "PNG" ) img.close()
def main(): # g13: parameter setting ------------------- ''' posemodel is trained_checkpoints/linemod/pose_model_9_0.01310166542980859.pth refine model is trained_checkpoints/linemod/pose_refine_model_493_0.006761023565178073.pth ''' objlist = [1, 2, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15] knn = KNearestNeighbor(1) opt.dataset ='linemod' opt.dataset_root = './datasets/linemod/Linemod_preprocessed' estimator_path = 'trained_checkpoints/linemod/pose_model_9_0.01310166542980859.pth' refiner_path = 'trained_checkpoints/linemod/pose_refine_model_493_0.006761023565178073.pth' opt.model = estimator_path opt.refine_model = refiner_path dataset_config_dir = 'datasets/linemod/dataset_config' output_result_dir = 'experiments/eval_result/linemod' opt.refine_start = True bs = 1 #fixed because of the default setting in torch.utils.data.DataLoader opt.iteration = 2 #default is 4 in eval_linemod.py t1_start = True t1_idx = 0 t1_total_eval_num = 3 t2_start = False t2_target_list = [22, 30, 172, 187, 267, 363, 410, 471, 472, 605, 644, 712, 1046, 1116, 1129, 1135, 1263] #t2_target_list = [0, 1] axis_range = 0.1 # the length of X, Y, and Z axis in 3D vimg_dir = 'verify_img' diameter = [] meta_file = open('{0}/models_info.yml'.format(dataset_config_dir), 'r') meta_d = yaml.load(meta_file) for obj in objlist: diameter.append(meta_d[obj]['diameter'] / 1000.0 * 0.1) print(diameter) if not os.path.exists(vimg_dir): os.makedirs(vimg_dir) #------------------------------------------- if opt.dataset == 'ycb': opt.num_objects = 21 #number of object classes in the dataset opt.num_points = 1000 #number of points on the input pointcloud opt.outf = 'trained_models/ycb' #folder to save trained models opt.log_dir = 'experiments/logs/ycb' #folder to save logs opt.repeat_epoch = 1 #number of repeat times for one epoch training elif opt.dataset == 'linemod': opt.num_objects = 13 opt.num_points = 500 opt.outf = 'trained_models/linemod' opt.log_dir = 'experiments/logs/linemod' opt.repeat_epoch = 20 else: print('Unknown dataset') return estimator = PoseNet(num_points = opt.num_points, num_obj = opt.num_objects) estimator.cuda() refiner = PoseRefineNet(num_points = opt.num_points, num_obj = opt.num_objects) refiner.cuda() estimator.load_state_dict(torch.load(estimator_path)) refiner.load_state_dict(torch.load(refiner_path)) opt.refine_start = True test_dataset = PoseDataset_linemod('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) opt.sym_list = test_dataset.get_sym_list() opt.num_points_mesh = test_dataset.get_num_points_mesh() print('>>>>>>>>----------Dataset loaded!---------<<<<<<<<\n\ length of the testing set: {0}\nnumber of sample points on mesh: {1}\n\ symmetry object list: {2}'\ .format( len(test_dataset), opt.num_points_mesh, opt.sym_list)) #load pytorch model estimator.eval() refiner.eval() criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list) fw = open('{0}/t1_eval_result_logs.txt'.format(output_result_dir), 'w') #Pose estimation for j, data in enumerate(testdataloader, 0): # g13: modify this part for evaluation target-------------------- if t1_start and j == t1_total_eval_num: break if t2_start and not (j in t2_target_list): continue #---------------------------------------------------------------- points, choose, img, target, model_points, idx = data if len(points.size()) == 2: print('No.{0} NOT Pass! Lost detection!'.format(j)) fw.write('No.{0} NOT Pass! Lost detection!\n'.format(j)) continue points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator(img, points, choose, idx) _, dis, new_points, new_target = criterion(pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) #if opt.refine_start: #iterative poserefinement # for ite in range(0, opt.iteration): # pred_r, pred_t = refiner(new_points, emb, idx) # dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_target, model_points, idx, new_points) pred_r = pred_r / torch.norm(pred_r, dim=2).view(1, opt.num_points, 1) pred_c = pred_c.view(bs, opt.num_points) how_max, which_max = torch.max(pred_c, 1) pred_t = pred_t.view(bs * opt.num_points, 1, 3) my_r = pred_r[0][which_max[0]].view(-1).cpu().data.numpy() my_t = (points.view(bs * opt.num_points, 1, 3) + pred_t)[which_max[0]].view(-1).cpu().data.numpy() my_pred = np.append(my_r, my_t) for ite in range(0, opt.iteration): T = Variable(torch.from_numpy(my_t.astype(np.float32))).cuda().view(1, 3).repeat(opt.num_points, 1).contiguous().view(1, opt.num_points, 3) my_mat = quaternion_matrix(my_r) R = Variable(torch.from_numpy(my_mat[:3, :3].astype(np.float32))).cuda().view(1, 3, 3) my_mat[0:3, 3] = my_t new_points = torch.bmm((points - T), R).contiguous() pred_r, pred_t = refiner(new_points, emb, idx) pred_r = pred_r.view(1, 1, -1) pred_r = pred_r / (torch.norm(pred_r, dim=2).view(1, 1, 1)) my_r_2 = pred_r.view(-1).cpu().data.numpy() my_t_2 = pred_t.view(-1).cpu().data.numpy() my_mat_2 = quaternion_matrix(my_r_2) my_mat_2[0:3, 3] = my_t_2 my_mat_final = np.dot(my_mat, my_mat_2) my_r_final = copy.deepcopy(my_mat_final) my_r_final[0:3, 3] = 0 my_r_final = quaternion_from_matrix(my_r_final, True) my_t_final = np.array([my_mat_final[0][3], my_mat_final[1][3], my_mat_final[2][3]]) my_pred = np.append(my_r_final, my_t_final) my_r = my_r_final my_t = my_t_final # Here 'my_pred' is the final pose estimation result after refinement ('my_r': quaternion, 'my_t': translation) #g13: checking the dis value success_count = [0 for i in range(opt.num_objects)] num_count = [0 for i in range(opt.num_objects)] model_points = model_points[0].cpu().detach().numpy() my_r = quaternion_matrix(my_r)[:3, :3] pred = np.dot(model_points, my_r.T) + my_t target = target[0].cpu().detach().numpy() if idx[0].item() in opt.sym_list: pred = torch.from_numpy(pred.astype(np.float32)).cuda().transpose(1, 0).contiguous() target = torch.from_numpy(target.astype(np.float32)).cuda().transpose(1, 0).contiguous() inds = knn(target.unsqueeze(0), pred.unsqueeze(0)) target = torch.index_select(target, 1, inds.view(-1) - 1) dis = torch.mean(torch.norm((pred.transpose(1, 0) - target.transpose(1, 0)), dim=1), dim=0).item() else: dis = np.mean(np.linalg.norm(pred - target, axis=1)) if dis < diameter[idx[0].item()]: success_count[idx[0].item()] += 1 print('No.{0} Pass! Distance: {1}'.format(j, dis)) fw.write('No.{0} Pass! Distance: {1}\n'.format(j, dis)) else: print('No.{0} NOT Pass! Distance: {1}'.format(j, dis)) fw.write('No.{0} NOT Pass! Distance: {1}\n'.format(j, dis)) num_count[idx[0].item()] += 1 # g13: start drawing pose on image------------------------------------ # pick up image print('{0}:\nmy_r is {1}\nmy_t is {2}\ndis:{3}'.format(j, my_r, my_t, dis.item())) print("index {0}: {1}".format(j, test_dataset.list_rgb[j])) img = Image.open(test_dataset.list_rgb[j]) # pick up center position by bbox meta_file = open('{0}/data/{1}/gt.yml'.format(opt.dataset_root, '%02d' % test_dataset.list_obj[j]), 'r') meta = {} meta = yaml.load(meta_file) which_item = test_dataset.list_rank[j] which_obj = test_dataset.list_obj[j] which_dict = 0 dict_leng = len(meta[which_item]) #print('get meta[{0}][{1}][obj_bb]'.format(which_item, which_obj)) k_idx = 0 while 1: if meta[which_item][k_idx]['obj_id'] == which_obj: which_dict = k_idx break k_idx = k_idx+1 bbx = meta[which_item][which_dict]['obj_bb'] draw = ImageDraw.Draw(img) # draw box (ensure this is the right object) draw.line((bbx[0],bbx[1], bbx[0], bbx[1]+bbx[3]), fill=(255,0,0), width=5) draw.line((bbx[0],bbx[1], bbx[0]+bbx[2], bbx[1]), fill=(255,0,0), width=5) draw.line((bbx[0],bbx[1]+bbx[3], bbx[0]+bbx[2], bbx[1]+bbx[3]), fill=(255,0,0), width=5) draw.line((bbx[0]+bbx[2],bbx[1], bbx[0]+bbx[2], bbx[1]+bbx[3]), fill=(255,0,0), width=5) #get center c_x = bbx[0]+int(bbx[2]/2) c_y = bbx[1]+int(bbx[3]/2) draw.point((c_x,c_y), fill=(255,255,0)) print('center:({0},{1})'.format(c_x, c_y)) #get the 3D position of center cam_intrinsic = np.zeros((3,3)) cam_intrinsic.itemset(0, test_dataset.cam_fx) cam_intrinsic.itemset(4, test_dataset.cam_fy) cam_intrinsic.itemset(2, test_dataset.cam_cx) cam_intrinsic.itemset(5, test_dataset.cam_cy) cam_intrinsic.itemset(8, 1) cam_extrinsic = my_mat_final[0:3, :] cam2d_3d = np.matmul(cam_intrinsic, cam_extrinsic) cen_3d = np.matmul(np.linalg.pinv(cam2d_3d), [[c_x],[c_y],[1]]) # replace img.show() with plt.imshow(img) #transpose three 3D axis point into 2D x_3d = cen_3d + [[axis_range],[0],[0],[0]] y_3d = cen_3d + [[0],[axis_range],[0],[0]] z_3d = cen_3d + [[0],[0],[axis_range],[0]] x_2d = np.matmul(cam2d_3d, x_3d) y_2d = np.matmul(cam2d_3d, y_3d) z_2d = np.matmul(cam2d_3d, z_3d) #draw the axis on 2D draw.line((c_x, c_y, x_2d[0], x_2d[1]), fill=(255,255,0), width=5) draw.line((c_x, c_y, y_2d[0], y_2d[1]), fill=(0,255,0), width=5) draw.line((c_x, c_y, z_2d[0], z_2d[1]), fill=(0,0,255), width=5) #g13: draw the estimate pred obj for pti in pred: pti.transpose() pti_2d = np.matmul(cam_intrinsic, pti) #print('({0},{1})\n'.format(int(pti_2d[0]),int(pti_2d[1]))) draw.point([int(pti_2d[0]),int(pti_2d[1])], fill=(255,255,0)) #g13: show image #img.show() #save file under file img_file_name = '{0}/batch{1}_pred_obj{2}_pic{3}.png'.format(vimg_dir, j, test_dataset.list_obj[j], which_item) img.save( img_file_name, "PNG" ) img.close() # plot ground true ---------------------------- img = Image.open(test_dataset.list_rgb[j]) draw = ImageDraw.Draw(img) draw.line((bbx[0],bbx[1], bbx[0], bbx[1]+bbx[3]), fill=(255,0,0), width=5) draw.line((bbx[0],bbx[1], bbx[0]+bbx[2], bbx[1]), fill=(255,0,0), width=5) draw.line((bbx[0],bbx[1]+bbx[3], bbx[0]+bbx[2], bbx[1]+bbx[3]), fill=(255,0,0), width=5) draw.line((bbx[0]+bbx[2],bbx[1], bbx[0]+bbx[2], bbx[1]+bbx[3]), fill=(255,0,0), width=5) target_r = np.resize(np.array(meta[which_item][k_idx]['cam_R_m2c']), (3, 3)) target_t = np.array(meta[which_item][k_idx]['cam_t_m2c']) target_t = target_t[np.newaxis, :] cam_extrinsic_GT = np.concatenate((target_r, target_t.T), axis=1) #get center 3D cam2d_3d_GT = np.matmul(cam_intrinsic, cam_extrinsic_GT) cen_3d_GT = np.matmul(np.linalg.pinv(cam2d_3d_GT), [[c_x],[c_y],[1]]) #transpose three 3D axis point into 2D x_3d = cen_3d_GT + [[axis_range],[0],[0],[0]] y_3d = cen_3d_GT + [[0],[axis_range],[0],[0]] z_3d = cen_3d_GT + [[0],[0],[axis_range],[0]] x_2d = np.matmul(cam2d_3d_GT, x_3d) y_2d = np.matmul(cam2d_3d_GT, y_3d) z_2d = np.matmul(cam2d_3d_GT, z_3d) #draw the axis on 2D draw.line((c_x, c_y, x_2d[0], x_2d[1]), fill=(255,255,0), width=5) draw.line((c_x, c_y, y_2d[0], y_2d[1]), fill=(0,255,0), width=5) draw.line((c_x, c_y, z_2d[0], z_2d[1]), fill=(0,0,255), width=5) print('pred:\n{0}\nGT:\n{1}\n'.format(cam_extrinsic,cam_extrinsic_GT)) print('pred 3D:{0}\nGT 3D:{1}\n'.format(cen_3d, cen_3d_GT)) img_file_name = '{0}/batch{1}_pred_obj{2}_pic{3}_gt.png'.format(vimg_dir, j, test_dataset.list_obj[j], which_item) img.save( img_file_name, "PNG" ) img.close() meta_file.close() print('\nplot_result_img.py completed the task\n')
def main(): if opt.dataset == 'linemod': opt.num_obj = 1 opt.list_obj = [1, 2, 4, 5, 6, 8, 9, 10, 11, 12, 13, 14, 15] opt.occ_list_obj = [1, 5, 6, 8, 9, 10, 11, 12] opt.list_name = ['ape', 'benchvise', 'cam', 'can', 'cat', 'driller', 'duck', 'eggbox', 'glue', 'holepuncher', 'iron', 'lamp', 'phone'] obj_name = opt.list_name[opt.list_obj.index(opt.obj_id)] opt.sym_list = [10, 11] opt.num_points = 500 meta_file = open('{0}/models/models_info.yml'.format(opt.dataset_root), 'r') meta = yaml.load(meta_file) diameter = meta[opt.obj_id]['diameter'] / 1000.0 * 0.1 if opt.render: opt.repeat_num = 1 elif opt.fuse: opt.repeat_num = 1 else: opt.repeat_num = 5 writer = SummaryWriter('experiments/runs/linemod/{}{}'.format(obj_name, opt.experiment_name)) opt.outf = 'trained_models/linemod/{}{}'.format(obj_name, opt.experiment_name) opt.log_dir = 'experiments/logs/linemod/{}{}'.format(obj_name, opt.experiment_name) if not os.path.exists(opt.outf): os.mkdir(opt.outf) if not os.path.exists(opt.log_dir): os.mkdir(opt.log_dir) else: print('Unknown dataset') return estimator = PoseNet(num_points = opt.num_points, num_vote = 9, num_obj = opt.num_obj) estimator.cuda() refiner = PoseRefineNet(num_points = opt.num_points, num_obj = opt.num_obj) refiner.cuda() if opt.resume_posenet != '': estimator.load_state_dict(torch.load('{0}/{1}'.format(opt.outf, opt.resume_posenet))) if opt.resume_refinenet != '': refiner.load_state_dict(torch.load('{0}/{1}'.format(opt.outf, opt.resume_refinenet))) opt.refine_start = True opt.lr = opt.lr_refine opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) else: opt.refine_start = False optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) dataset = PoseDataset_linemod('train', opt.num_points, opt.dataset_root, opt.real, opt.render, opt.fuse, opt.obj_id) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) test_dataset = PoseDataset_linemod('test', opt.num_points, opt.dataset_root, True, False, False, opt.obj_id) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) print('>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}'.format(len(dataset), len(test_dataset), opt.num_points)) if opt.obj_id in opt.occ_list_obj: occ_test_dataset = PoseDataset_occ('test', opt.num_points, opt.occ_dataset_root, opt.obj_id) occtestdataloader = torch.utils.data.DataLoader(occ_test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) print('length of the occ testing set: {}'.format(len(occ_test_dataset))) criterion = Loss(opt.num_points, opt.sym_list) criterion_refine = Loss_refine(opt.num_points, opt.sym_list) best_test = np.Inf if opt.start_epoch == 1: for log in os.listdir(opt.log_dir): os.remove(os.path.join(opt.log_dir, log)) st_time = time.time() train_scalar = 0 for epoch in range(opt.start_epoch, opt.nepoch): logger = setup_logger('epoch%d' % epoch, os.path.join(opt.log_dir, 'epoch_%d_log.txt' % epoch)) logger.info('Train time {0}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Training started')) train_count = 0 train_loss_avg = 0.0 train_loss = 0.0 train_dis_avg = 0.0 train_dis = 0.0 if opt.refine_start: estimator.eval() refiner.train() else: estimator.train() optimizer.zero_grad() for rep in range(opt.repeat_num): for i, data in enumerate(dataloader, 0): points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = data if len(points.size()) == 2: print('pass') continue points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = points.cuda(), choose.cuda(), img.cuda(), target.cuda(), model_points.cuda(), model_kp.cuda(), vertex_gt.cuda(), idx.cuda(), target_r.cuda(), target_t.cuda() vertex_pred, c_pred, emb = estimator(img, points, choose, idx) vertex_loss, pose_loss, dis, new_points, new_target = criterion(vertex_pred, vertex_gt, c_pred, points, target, model_points, model_kp, opt.obj_id, target_r, target_t) loss = 10 * vertex_loss + pose_loss if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_points, new_target, model_points, opt.obj_id) dis.backward() else: loss.backward() train_loss_avg += loss.item() train_loss += loss.item() train_dis_avg += dis.item() train_dis += dis.item() train_count += 1 train_scalar += 1 if train_count % opt.batch_size == 0: logger.info('Train time {0} Epoch {1} Batch {2} Frame {3} Avg_loss:{4} Avg_diss:{5}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, int(train_count / opt.batch_size), train_count, train_loss_avg / opt.batch_size, train_dis_avg / opt.batch_size)) writer.add_scalar('linemod training loss', train_loss_avg / opt.batch_size, train_scalar) writer.add_scalar('linemod training dis', train_dis_avg / opt.batch_size, train_scalar) optimizer.step() optimizer.zero_grad() train_loss_avg = 0 train_dis_avg = 0 if train_count != 0 and train_count % 1000 == 0: if opt.refine_start: torch.save(refiner.state_dict(), '{0}/pose_refine_model_current.pth'.format(opt.outf)) else: torch.save(estimator.state_dict(), '{0}/pose_model_current.pth'.format(opt.outf)) print('>>>>>>>>----------epoch {0} train finish---------<<<<<<<<'.format(epoch)) train_loss = train_loss / train_count train_dis = train_dis / train_count logger.info('Train time {0} Epoch {1} TRAIN FINISH Avg loss: {2} Avg dis: {3}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, train_loss, train_dis)) logger = setup_logger('epoch%d_test' % epoch, os.path.join(opt.log_dir, 'epoch_%d_test_log.txt' % epoch)) logger.info('Test time {0}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Testing started')) test_loss = 0.0 test_vertex_loss = 0.0 test_pose_loss = 0.0 test_dis = 0.0 test_count = 0 success_count = 0 estimator.eval() refiner.eval() for j, data in enumerate(testdataloader, 0): points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = data if len(points.size()) == 2: logger.info('Test time {0} Lost detection!'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)))) continue points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = points.cuda(), choose.cuda(), img.cuda(), target.cuda(), model_points.cuda(), model_kp.cuda(), vertex_gt.cuda(), idx.cuda(), target_r.cuda(), target_t.cuda() vertex_pred, c_pred, emb = estimator(img, points, choose, idx) vertex_loss, pose_loss, dis, new_points, new_target = criterion(vertex_pred, vertex_gt, c_pred, points, target, model_points, model_kp, opt.obj_id, target_r, target_t) loss = 10 * vertex_loss + pose_loss if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_points, new_target, model_points, opt.obj_id) test_loss += loss.item() test_vertex_loss += vertex_loss.item() test_pose_loss += pose_loss.item() test_dis += dis.item() logger.info('Test time {0} Test Frame No.{1} loss:{2} dis:{3}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), test_count, loss, dis)) if dis.item() < diameter: success_count += 1 test_count += 1 test_loss = test_loss / test_count test_vertex_loss = test_vertex_loss / test_count test_pose_loss = test_pose_loss / test_count test_dis = test_dis / test_count success_rate = float(success_count) / test_count logger.info('Test time {0} Epoch {1} TEST FINISH Avg loss: {2} Avg dis: {3} Success rate: {4}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, test_loss, test_dis, success_rate)) writer.add_scalar('linemod test loss', test_loss, epoch) writer.add_scalar('linemod test vertex loss', test_vertex_loss, epoch) writer.add_scalar('linemod test pose loss', test_pose_loss, epoch) writer.add_scalar('linemod test dis', test_dis, epoch) writer.add_scalar('linemod success rate', success_rate, epoch) writer.add_scalar('lr', optimizer.param_groups[0]['lr'], epoch) if test_dis <= best_test: best_test = test_dis if opt.refine_start: torch.save(refiner.state_dict(), '{0}/pose_refine_model_{1}_{2}.pth'.format(opt.outf, epoch, test_dis)) else: torch.save(estimator.state_dict(), '{0}/pose_model_{1}_{2}.pth'.format(opt.outf, epoch, test_dis)) print(epoch, '>>>>>>>>----------MODEL SAVED---------<<<<<<<<') if opt.obj_id in opt.occ_list_obj: logger = setup_logger('epoch%d_occ_test' % epoch, os.path.join(opt.log_dir, 'epoch_%d_occ_test_log.txt' % epoch)) logger.info('Occ test time {0}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Testing started')) occ_test_dis = 0.0 occ_test_count = 0 occ_success_count = 0 estimator.eval() refiner.eval() for j, data in enumerate(occtestdataloader, 0): points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = data if len(points.size()) == 2: logger.info('Occ test time {0} Lost detection!'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)))) continue points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = points.cuda(), choose.cuda(), img.cuda(), target.cuda(), model_points.cuda(), model_kp.cuda(), vertex_gt.cuda(), idx.cuda(), target_r.cuda(), target_t.cuda() vertex_pred, c_pred, emb = estimator(img, points, choose, idx) vertex_loss, pose_loss, dis, new_points, new_target = criterion(vertex_pred, vertex_gt, c_pred, points, target, model_points, model_kp, opt.obj_id, target_r, target_t) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_points, new_target, model_points, opt.obj_id) occ_test_dis += dis.item() logger.info('Occ test time {0} Test Frame No.{1} dis:{2}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), occ_test_count, dis)) if dis.item() < diameter: occ_success_count += 1 occ_test_count += 1 occ_test_dis = occ_test_dis / occ_test_count occ_success_rate = float(occ_success_count) / occ_test_count logger.info('Occ test time {0} Epoch {1} TEST FINISH Avg dis: {2} Success rate: {3}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, occ_test_dis, occ_success_rate)) writer.add_scalar('occ test dis', occ_test_dis, epoch) writer.add_scalar('occ success rate', occ_success_rate, epoch) if best_test < opt.refine_margin and not opt.refine_start: opt.refine_start = True opt.lr = opt.lr_refine opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) print('>>>>>>>>----------Refine started---------<<<<<<<<') writer.close()
def main(): if opt.dataset == 'ycb': opt.num_obj = 21 opt.sym_list = [12, 15, 18, 19, 20] opt.num_points = 1000 writer = SummaryWriter('experiments/runs/ycb/{0}'.format(opt.experiment_name)) opt.outf = 'trained_models/ycb/{0}'.format(opt.experiment_name) opt.log_dir = 'experiments/logs/ycb/{0}'.format(opt.experiment_name) opt.repeat_num = 1 if not os.path.exists(opt.outf): os.mkdir(opt.outf) if not os.path.exists(opt.log_dir): os.mkdir(opt.log_dir) else: print('Unknown dataset') return estimator = PoseNet(num_points = opt.num_points, num_vote = 9, num_obj = opt.num_obj) estimator.cuda() refiner = PoseRefineNet(num_points = opt.num_points, num_obj = opt.num_obj) refiner.cuda() if opt.resume_posenet != '': estimator.load_state_dict(torch.load('{0}/{1}'.format(opt.outf, opt.resume_posenet))) if opt.resume_refinenet != '': refiner.load_state_dict(torch.load('{0}/{1}'.format(opt.outf, opt.resume_refinenet))) opt.refine_start = True opt.lr = opt.lr_refine opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) else: opt.refine_start = False optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) dataset = PoseDataset_ycb('train', opt.num_points, True, opt.dataset_root) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) test_dataset = PoseDataset_ycb('test', opt.num_points, False, opt.dataset_root) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) print('>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}'.format(len(dataset), len(test_dataset), opt.num_points)) criterion = Loss(opt.num_points, opt.sym_list) criterion_refine = Loss_refine(opt.num_points, opt.sym_list) best_test = np.Inf if opt.start_epoch == 1: for log in os.listdir(opt.log_dir): os.remove(os.path.join(opt.log_dir, log)) st_time = time.time() train_scalar = 0 for epoch in range(opt.start_epoch, opt.nepoch): logger = setup_logger('epoch%d' % epoch, os.path.join(opt.log_dir, 'epoch_%d_log.txt' % epoch)) logger.info('Train time {0}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Training started')) train_count = 0 train_loss_avg = 0.0 train_loss = 0.0 train_dis_avg = 0.0 train_dis = 0.0 if opt.refine_start: estimator.eval() refiner.train() else: estimator.train() optimizer.zero_grad() for rep in range(opt.repeat_num): for i, data in enumerate(dataloader, 0): points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = data points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = points.cuda(), choose.cuda(), img.cuda(), target.cuda(), model_points.cuda(), model_kp.cuda(), vertex_gt.cuda(), idx.cuda(), target_r.cuda(), target_t.cuda() vertex_pred, c_pred, emb = estimator(img, points, choose, idx) vertex_loss, pose_loss, dis, new_points, new_target = criterion(vertex_pred, vertex_gt, c_pred, points, target, model_points, model_kp, idx, target_r, target_t) loss = 10 * vertex_loss + pose_loss if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_points, new_target, model_points, idx) dis.backward() else: loss.backward() train_loss_avg += loss.item() train_loss += loss.item() train_dis_avg += dis.item() train_dis += dis.item() train_count += 1 train_scalar += 1 if train_count % opt.batch_size == 0: logger.info('Train time {0} Epoch {1} Batch {2} Frame {3} Avg_loss:{4} Avg_diss:{5}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, int(train_count / opt.batch_size), train_count, train_loss_avg / opt.batch_size, train_dis_avg / opt.batch_size)) writer.add_scalar('ycb training loss', train_loss_avg / opt.batch_size, train_scalar) writer.add_scalar('ycb training dis', train_dis_avg / opt.batch_size, train_scalar) optimizer.step() optimizer.zero_grad() train_loss_avg = 0 train_dis_avg = 0 if train_count != 0 and train_count % 1000 == 0: if opt.refine_start: torch.save(refiner.state_dict(), '{0}/pose_refine_model_current.pth'.format(opt.outf)) else: torch.save(estimator.state_dict(), '{0}/pose_model_current.pth'.format(opt.outf)) print('>>>>>>>>----------epoch {0} train finish---------<<<<<<<<'.format(epoch)) train_loss = train_loss / train_count train_dis = train_dis / train_count logger.info('Train time {0} Epoch {1} TRAIN FINISH Avg loss: {2} Avg dis: {3}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, train_loss, train_dis)) logger = setup_logger('epoch%d_test' % epoch, os.path.join(opt.log_dir, 'epoch_%d_test_log.txt' % epoch)) logger.info('Test time {0}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Testing started')) test_loss = 0.0 test_vertex_loss = 0.0 test_pose_loss = 0.0 test_dis = 0.0 test_count = 0 success_count = 0 estimator.eval() refiner.eval() for j, data in enumerate(testdataloader, 0): points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = data points, choose, img, target, model_points, model_kp, vertex_gt, idx, target_r, target_t = points.cuda(), choose.cuda(), img.cuda(), target.cuda(), model_points.cuda(), model_kp.cuda(), vertex_gt.cuda(), idx.cuda(), target_r.cuda(), target_t.cuda() vertex_pred, c_pred, emb = estimator(img, points, choose, idx) vertex_loss, pose_loss, dis, new_points, new_target = criterion(vertex_pred, vertex_gt, c_pred, points, target, model_points, model_kp, idx, target_r, target_t) loss = 10 * vertex_loss + pose_loss if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine(pred_r, pred_t, new_points, new_target, model_points, idx) test_loss += loss.item() test_vertex_loss += vertex_loss.item() test_pose_loss += pose_loss.item() test_dis += dis.item() logger.info('Test time {0} Test Frame No.{1} loss:{2} dis:{3}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), test_count, loss, dis)) test_count += 1 if dis.item() < 0.02: success_count += 1 test_loss = test_loss / test_count test_vertex_loss = test_vertex_loss / test_count test_pose_loss = test_pose_loss / test_count test_dis = test_dis / test_count logger.info('Test time {0} Epoch {1} TEST FINISH Avg loss: {2} Avg dis: {3}'.format(time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, test_loss, test_dis)) logger.info('Success rate: {}'.format(float(success_count) / test_count)) writer.add_scalar('ycb test loss', test_loss, epoch) writer.add_scalar('ycb test vertex loss', test_vertex_loss, epoch) writer.add_scalar('ycb test pose loss', test_pose_loss, epoch) writer.add_scalar('ycb test dis', test_dis, epoch) writer.add_scalar('ycb success rate', float(success_count) / test_count, epoch) writer.add_scalar('lr', optimizer.param_groups[0]['lr'], epoch) if test_dis <= best_test: best_test = test_dis if opt.refine_start: torch.save(refiner.state_dict(), '{0}/pose_refine_model_{1}_{2}.pth'.format(opt.outf, epoch, test_dis)) else: torch.save(estimator.state_dict(), '{0}/pose_model_{1}_{2}.pth'.format(opt.outf, epoch, test_dis)) print(epoch, '>>>>>>>>----------MODEL SAVED---------<<<<<<<<') if best_test < opt.refine_margin and not opt.refine_start: opt.refine_start = True opt.lr = opt.lr_refine opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) print('>>>>>>>>----------Refine started---------<<<<<<<<') writer.close()
def main(): opt.manualSeed = random.randint(1, 10000) random.seed(opt.manualSeed) torch.manual_seed(opt.manualSeed) if opt.dataset == 'linemod': opt.num_objects = 13 opt.num_points = 500 opt.outf = 'trained_models/linemod' opt.log_dir = 'experiments/logs/linemod' output_results = 'check_linemod.txt' opt.repeat_epoch = 20 elif opt.dataset == 'ycb': opt.num_objects = 21 #number of object classes in the dataset opt.num_points = 1000 #number of points on the input pointcloud opt.outf = 'trained_models/ycb' #folder to save trained models opt.log_dir = 'experiments/logs/ycb' #folder to save logs opt.repeat_epoch = 1 #number of repeat times for one epoch training elif opt.dataset == 'ycb-syn': opt.num_objects = 31 # number of object classes in the dataset opt.num_points = 1000 # number of points on the input pointcloud opt.dataset_root = '/data/Akeaveny/Datasets/ycb_syn' opt.outf = 'trained_models/ycb_syn/ycb_syn2' # folder to save trained models opt.log_dir = 'experiments/logs/ycb_syn/ycb_syn2' # folder to save logs output_results = 'check_ycb_syn.txt' opt.w = 0.05 opt.refine_margin = 0.01 elif opt.dataset == 'arl': opt.num_objects = 10 # number of object classes in the dataset opt.num_points = 1000 # number of points on the input pointcloud opt.dataset_root = '/data/Akeaveny/Datasets/arl_dataset' opt.outf = 'trained_models/arl/clutter/arl_finetune_syn_2' # folder to save trained models opt.log_dir = '/home/akeaveny/catkin_ws/src/object-rpe-ak/DenseFusion/experiments/logs/arl/clutter/arl_finetune_syn_2' # folder to save logs output_results = 'check_arl_syn.txt' opt.nepoch = 750 opt.w = 0.05 opt.refine_margin = 0.0045 # TODO opt.repeat_epoch = 20 opt.start_epoch = 0 opt.resume_posenet = 'pose_model_1_0.012397416144377301.pth' opt.resume_refinenet = 'pose_refine_model_153_0.004032851301599294.pth' elif opt.dataset == 'arl1': opt.num_objects = 5 # number of object classes in the dataset opt.num_points = 1000 # number of points on the input pointcloud opt.dataset_root = '/data/Akeaveny/Datasets/arl_dataset' opt.outf = 'trained_models/arl1/clutter/arl_real_2' # folder to save trained models opt.log_dir = '/home/akeaveny/catkin_ws/src/object-rpe-ak/DenseFusion/experiments/logs/arl1/clutter/arl_real_2' # folder to save logs output_results = 'check_arl_syn.txt' opt.nepoch = 750 opt.w = 0.05 opt.refine_margin = 0.015 # opt.start_epoch = 120 # opt.resume_posenet = 'pose_model_current.pth' # opt.resume_refinenet = 'pose_refine_model_115_0.008727498716640046.pth' elif opt.dataset == 'elevator': opt.num_objects = 1 # number of object classes in the dataset opt.num_points = 1000 # number of points on the input pointcloud opt.dataset_root = '/data/Akeaveny/Datasets/elevator_dataset' opt.outf = 'trained_models/elevator/elevator_2' # folder to save trained models opt.log_dir = '/home/akeaveny/catkin_ws/src/object-rpe-ak/DenseFusion/experiments/logs/elevator/elevator_2' # folder to save logs output_results = 'check_arl_syn.txt' opt.nepoch = 750 opt.w = 0.05 opt.refine_margin = 0.015 opt.nepoch = 750 opt.w = 0.05 opt.refine_margin = 0.015 # TODO opt.repeat_epoch = 40 # opt.start_epoch = 47 # opt.resume_posenet = 'pose_model_current.pth' # opt.resume_refinenet = 'pose_refine_model_46_0.007581770288279472.pth' else: print('Unknown dataset') return estimator = PoseNet(num_points=opt.num_points, num_obj=opt.num_objects) estimator.cuda() refiner = PoseRefineNet(num_points=opt.num_points, num_obj=opt.num_objects) refiner.cuda() if opt.resume_posenet != '': estimator.load_state_dict( torch.load('{0}/{1}'.format(opt.outf, opt.resume_posenet))) if opt.resume_refinenet != '': refiner.load_state_dict( torch.load('{0}/{1}'.format(opt.outf, opt.resume_refinenet))) opt.refine_start = False opt.decay_start = False opt.lr *= opt.lr_rate opt.w *= opt.w_rate opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) else: opt.refine_start = False opt.decay_start = False optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) if opt.dataset == 'ycb': dataset = PoseDataset_ycb('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'linemod': dataset = PoseDataset_linemod('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'ycb-syn': dataset = PoseDataset_ycb_syn('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'arl': dataset = PoseDataset_arl('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'arl1': dataset = PoseDataset_arl1('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'elevator': dataset = PoseDataset_elevator('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) if opt.dataset == 'ycb': test_dataset = PoseDataset_ycb('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'linemod': test_dataset = PoseDataset_linemod('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'ycb-syn': test_dataset = PoseDataset_ycb_syn('test', opt.num_points, True, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'arl': test_dataset = PoseDataset_arl('test', opt.num_points, True, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'arl1': test_dataset = PoseDataset_arl1('test', opt.num_points, True, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'elevator': test_dataset = PoseDataset_elevator('test', opt.num_points, True, opt.dataset_root, 0.0, opt.refine_start) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) opt.sym_list = dataset.get_sym_list() opt.num_points_mesh = dataset.get_num_points_mesh() print( '>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}\nsymmetry object list: {3}' .format(len(dataset), len(test_dataset), opt.num_points_mesh, opt.sym_list)) criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list) best_test = np.Inf if opt.start_epoch == 1: for log in os.listdir(opt.log_dir): os.remove(os.path.join(opt.log_dir, log)) st_time = time.time() ###################### ###################### # TODO (ak): set up tensor board # if not os.path.exists(opt.log_dir): # os.makedirs(opt.log_dir) # # writer = SummaryWriter(opt.log_dir) ###################### ###################### for epoch in range(opt.start_epoch, opt.nepoch): logger = setup_logger( 'epoch%d' % epoch, os.path.join(opt.log_dir, 'epoch_%d_log.txt' % epoch)) logger.info('Train time {0}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Training started')) train_count = 0 train_dis_avg = 0.0 if opt.refine_start: estimator.eval() refiner.train() else: estimator.train() optimizer.zero_grad() for rep in range(opt.repeat_epoch): ################## # train ################## for i, data in enumerate(dataloader, 0): points, choose, img, target, model_points, idx = data # TODO: txt file # fw = open(test_folder + output_results, 'w') # fw.write('Points\n{0}\n\nchoose\n{1}\n\nimg\n{2}\n\ntarget\n{3}\n\nmodel_points\n{4}'.format(points, choose, img, target, model_points)) # fw.close() points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator( img, points, choose, idx) loss, dis, new_points, new_target = criterion( pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine( pred_r, pred_t, new_target, model_points, idx, new_points) dis.backward() else: loss.backward() train_dis_avg += dis.item() train_count += 1 if train_count % opt.batch_size == 0: logger.info( 'Train time {} Epoch {} Batch {} Frame {}/{} Avg_dis: {:.2f} [cm]' .format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, int(train_count / opt.batch_size), train_count, len(dataset.list), train_dis_avg / opt.batch_size * 100)) optimizer.step() optimizer.zero_grad() # TODO: tensorboard # if train_count != 0 and train_count % 250 == 0: # scalar_info = {'loss': loss.item(), # 'dis': train_dis_avg / opt.batch_size} # for key, val in scalar_info.items(): # writer.add_scalar(key, val, train_count) train_dis_avg = 0 if train_count != 0 and train_count % 1000 == 0: if opt.refine_start: torch.save( refiner.state_dict(), '{0}/pose_refine_model_current.pth'.format( opt.outf)) else: torch.save( estimator.state_dict(), '{0}/pose_model_current.pth'.format(opt.outf)) # TODO: tensorboard # scalar_info = {'loss': loss.item(), # 'dis': dis.item()} # for key, val in scalar_info.items(): # writer.add_scalar(key, val, train_count) print( '>>>>>>>>----------epoch {0} train finish---------<<<<<<<<'.format( epoch)) logger = setup_logger( 'epoch%d_test' % epoch, os.path.join(opt.log_dir, 'epoch_%d_test_log.txt' % epoch)) logger.info('Test time {0}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Testing started')) test_dis = 0.0 test_count = 0 estimator.eval() refiner.eval() for j, data in enumerate(testdataloader, 0): points, choose, img, target, model_points, idx = data points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator(img, points, choose, idx) _, dis, new_points, new_target = criterion(pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine( pred_r, pred_t, new_target, model_points, idx, new_points) test_dis += dis.item() logger.info('Test time {} Test Frame No.{} dis: {} [cm]'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), test_count, dis * 100)) test_count += 1 test_dis = test_dis / test_count logger.info( 'Test time {} Epoch {} TEST FINISH Avg dis: {} [cm]'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, test_dis * 100)) # TODO: tensorboard # scalar_info = {'test dis': test_dis} # for key, val in scalar_info.items(): # writer.add_scalar(key, val, train_count) if test_dis <= best_test: best_test = test_dis if opt.refine_start: torch.save( refiner.state_dict(), '{0}/pose_refine_model_{1}_{2}.pth'.format( opt.outf, epoch, test_dis)) else: torch.save( estimator.state_dict(), '{0}/pose_model_{1}_{2}.pth'.format( opt.outf, epoch, test_dis)) print(epoch, '>>>>>>>>----------BEST TEST MODEL SAVED---------<<<<<<<<') if best_test < opt.decay_margin and not opt.decay_start: opt.decay_start = True opt.lr *= opt.lr_rate opt.w *= opt.w_rate optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) if best_test < opt.refine_margin and not opt.refine_start: opt.refine_start = True opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) if opt.dataset == 'ycb': dataset = PoseDataset_ycb('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'linemod': dataset = PoseDataset_linemod('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'ycb-syn': dataset = PoseDataset_ycb_syn('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'arl': dataset = PoseDataset_arl('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'arl1': dataset = PoseDataset_arl1('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) elif opt.dataset == 'elevator': dataset = PoseDataset_elevator('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) if opt.dataset == 'ycb': test_dataset = PoseDataset_ycb('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'linemod': test_dataset = PoseDataset_linemod('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'ycb-syn': test_dataset = PoseDataset_ycb_syn('test', opt.num_points, True, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'arl': test_dataset = PoseDataset_arl('test', opt.num_points, True, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'arl1': test_dataset = PoseDataset_arl1('test', opt.num_points, True, opt.dataset_root, 0.0, opt.refine_start) elif opt.dataset == 'elevator': test_dataset = PoseDataset_elevator('test', opt.num_points, True, opt.dataset_root, 0.0, opt.refine_start) testdataloader = torch.utils.data.DataLoader( test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) opt.sym_list = dataset.get_sym_list() opt.num_points_mesh = dataset.get_num_points_mesh() print( '>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}\nsymmetry object list: {3}' .format(len(dataset), len(test_dataset), opt.num_points_mesh, opt.sym_list)) criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list)
def main(): opt.manualSeed = random.randint(1, 10000) random.seed(opt.manualSeed) torch.manual_seed(opt.manualSeed) opt.num_objects = 3 opt.num_points = 500 opt.outf = 'trained_models' opt.log_dir = 'experiments/logs' opt.repeat_epoch = 20 estimator = PoseNet(num_points=opt.num_points, num_obj=opt.num_objects) estimator.cuda() refiner = PoseRefineNet(num_points=opt.num_points, num_obj=opt.num_objects) refiner.cuda() if opt.resume_posenet != '': estimator.load_state_dict( torch.load('{0}/{1}'.format(opt.outf, opt.resume_posenet))) if opt.resume_refinenet != '': refiner.load_state_dict( torch.load('{0}/{1}'.format(opt.outf, opt.resume_refinenet))) opt.refine_start = True opt.decay_start = True opt.lr *= opt.lr_rate opt.w *= opt.w_rate opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) else: opt.refine_start = False opt.decay_start = False optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) dataset = PoseDataset('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) test_dataset = PoseDataset('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) testdataloader = torch.utils.data.DataLoader(test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) opt.sym_list = dataset.get_sym_list() opt.num_points_mesh = dataset.get_num_points_mesh() print( '>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}' .format(len(dataset), len(test_dataset), opt.num_points_mesh)) criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list) best_test = np.Inf if opt.start_epoch == 1: for log in os.listdir(opt.log_dir): os.remove(os.path.join(opt.log_dir, log)) st_time = time.time() for epoch in range(opt.start_epoch, opt.nepoch): logger = setup_logger( 'epoch%d' % epoch, os.path.join(opt.log_dir, 'epoch_%d_log.txt' % epoch)) logger.info('Train time {0}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Training started')) train_count = 0 train_dis_avg = 0.0 if opt.refine_start: estimator.eval() # affects dropout and batch normalization refiner.train() else: estimator.train() optimizer.zero_grad() for rep in range(opt.repeat_epoch): for i, data in enumerate(dataloader, 0): points, choose, img, target, model_points, idx = data #points ->torch.Size([500, 3]) ->在crop出来的像素区域随机选取500个点,利用相机内参结合深度值算出来的点云cloud #choose ->torch.Size([1, 500]) #img ->torch.Size([3, 80, 80]) #target ->torch.Size([500, 3]) ->真实模型上随机选取的mesh点进行ground truth pose变换后得到的点 #model_points ->torch.Size([500, 3]) ->真实模型上随机选取的mesh点在进行pose变换前的点 #idx ->torch.Size([1]) #tensor([4], device='cuda:0') #img和points对应rgb和点云信息,需要在网络内部fusion points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator( img, points, choose, idx) loss, dis, new_points, new_target = criterion( pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine( pred_r, pred_t, new_target, model_points, idx, new_points) dis.backward() else: loss.backward() train_dis_avg += dis.item() train_count += 1 if train_count % opt.batch_size == 0: logger.info( 'Train time {0} Epoch {1} Batch {2} Frame {3} Avg_dis:{4}' .format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, int(train_count / opt.batch_size), train_count, train_dis_avg / opt.batch_size)) optimizer.step() optimizer.zero_grad() train_dis_avg = 0 if train_count != 0 and train_count % 1000 == 0: if opt.refine_start: torch.save( refiner.state_dict(), '{0}/pose_refine_model_current.pth'.format( opt.outf)) else: torch.save( estimator.state_dict(), '{0}/pose_model_current.pth'.format(opt.outf)) print( '>>>>>>>>----------epoch {0} train finish---------<<<<<<<<'.format( epoch)) logger = setup_logger( 'epoch%d_test' % epoch, os.path.join(opt.log_dir, 'epoch_%d_test_log.txt' % epoch)) logger.info('Test time {0}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)) + ', ' + 'Testing started')) test_dis = 0.0 test_count = 0 estimator.eval() refiner.eval() for j, data in enumerate(testdataloader, 0): points, choose, img, target, model_points, idx = data points, choose, img, target, model_points, idx = Variable(points).cuda(), \ Variable(choose).cuda(), \ Variable(img).cuda(), \ Variable(target).cuda(), \ Variable(model_points).cuda(), \ Variable(idx).cuda() pred_r, pred_t, pred_c, emb = estimator(img, points, choose, idx) _, dis, new_points, new_target = criterion(pred_r, pred_t, pred_c, target, model_points, idx, points, opt.w, opt.refine_start) if opt.refine_start: for ite in range(0, opt.iteration): pred_r, pred_t = refiner(new_points, emb, idx) dis, new_points, new_target = criterion_refine( pred_r, pred_t, new_target, model_points, idx, new_points) test_dis += dis.item() logger.info('Test time {0} Test Frame No.{1} dis:{2}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), test_count, dis)) test_count += 1 test_dis = test_dis / test_count logger.info('Test time {0} Epoch {1} TEST FINISH Avg dis: {2}'.format( time.strftime("%Hh %Mm %Ss", time.gmtime(time.time() - st_time)), epoch, test_dis)) if test_dis <= best_test: best_test = test_dis if opt.refine_start: torch.save( refiner.state_dict(), '{0}/pose_refine_model_{1}_{2}.pth'.format( opt.outf, epoch, test_dis)) else: torch.save( estimator.state_dict(), '{0}/pose_model_{1}_{2}.pth'.format( opt.outf, epoch, test_dis)) print(epoch, '>>>>>>>>----------BEST TEST MODEL SAVED---------<<<<<<<<') if best_test < opt.decay_margin and not opt.decay_start: opt.decay_start = True opt.lr *= opt.lr_rate opt.w *= opt.w_rate optimizer = optim.Adam(estimator.parameters(), lr=opt.lr) if best_test < opt.refine_margin and not opt.refine_start: opt.refine_start = True opt.batch_size = int(opt.batch_size / opt.iteration) optimizer = optim.Adam(refiner.parameters(), lr=opt.lr) dataset = PoseDataset('train', opt.num_points, True, opt.dataset_root, opt.noise_trans, opt.refine_start) dataloader = torch.utils.data.DataLoader(dataset, batch_size=1, shuffle=True, num_workers=opt.workers) test_dataset = PoseDataset('test', opt.num_points, False, opt.dataset_root, 0.0, opt.refine_start) testdataloader = torch.utils.data.DataLoader( test_dataset, batch_size=1, shuffle=False, num_workers=opt.workers) opt.sym_list = dataset.get_sym_list() opt.num_points_mesh = dataset.get_num_points_mesh() print( '>>>>>>>>----------Dataset loaded!---------<<<<<<<<\nlength of the training set: {0}\nlength of the testing set: {1}\nnumber of sample points on mesh: {2}' .format(len(dataset), len(test_dataset), opt.num_points_mesh)) criterion = Loss(opt.num_points_mesh, opt.sym_list) criterion_refine = Loss_refine(opt.num_points_mesh, opt.sym_list)