def assess_acc_block(net, test_loader, gt_k=None, include_rgb=None, penultimate_features=False, contiguous_sz=None): total = 0 all = None all_targets = None for i, (imgs, targets) in enumerate(test_loader): imgs = Variable(sobel_process(imgs.cuda(), include_rgb)) with torch.no_grad(): x_out = net(imgs, penultimate_features=penultimate_features) bn, dlen = x_out.shape if all is None: all = np.zeros((len(test_loader) * bn, dlen)) all_targets = np.zeros(len(test_loader) * bn) all[total:(total + bn), :] = x_out.cpu().numpy() all_targets[total:(total + bn)] = targets.numpy() total += bn # 40000 all = all[:total, :] all_targets = all_targets[:total] num_orig, leftover = divmod(total, contiguous_sz) assert (leftover == 0) all = all.reshape((num_orig, contiguous_sz, dlen)) all = all.sum(axis=1, keepdims=False) / float(contiguous_sz) all_targets = all_targets.reshape((num_orig, contiguous_sz)) # sanity check all_targets_avg = all_targets.astype("int").sum(axis=1) / contiguous_sz all_targets = all_targets[:, 0].astype("int") assert (np.array_equal(all_targets_avg, all_targets)) preds = np.argmax(all, axis=1).astype("int") assert (preds.min() >= 0 and preds.max() < gt_k) assert (all_targets.min() >= 0 and all_targets.max() < gt_k) if not (preds.shape == all_targets.shape): print(preds.shape) print(all_targets.shape) assert (False) assert (preds.shape == (num_orig, )) correct = (preds == all_targets).sum() return correct / float(num_orig)
def get_dlen(net_features, dataloader, include_rgb=None, penultimate_features=False): for i, (imgs, _) in enumerate(dataloader): imgs = Variable(sobel_process(imgs.cuda(), include_rgb)).cpu() x_features = net_features(imgs, trunk_features=True, penultimate_features=penultimate_features) x_features = x_features.view(x_features.shape[0], -1) dlen = x_features.shape[1] break return dlen
def assess_acc(net, test_loader, gt_k=None, include_rgb=None, penultimate_features=False): correct = 0 total = 0 for i, (imgs, targets) in enumerate(test_loader): imgs = Variable(sobel_process(imgs.cuda(), include_rgb)) with torch.no_grad(): x_out = net(imgs, penultimate_features=penultimate_features) # bug fix!! preds = np.argmax(x_out.cpu().numpy(), axis=1).astype("int") targets = targets.numpy().astype("int") assert (preds.min() >= 0 and preds.max() < gt_k) assert (targets.min() >= 0 and targets.max() < gt_k) assert (preds.shape == targets.shape) correct += (preds == targets).sum() total += preds.shape[0] return correct / float(total)
elif imgs_dataloader_name == "train": imgs_dataloader = mapping_assignment_dataloader else: assert (False) print("length of imgs_dataloader %d" % len(imgs_dataloader)) next_img_ind = 0 for b_i, batch in enumerate(imgs_dataloader): orig_imgs, flat_targets, mask = batch orig_imgs, flat_targets, mask = \ orig_imgs.cuda(), flat_targets.numpy(), mask.numpy().astype(np.bool) if not config.no_sobel: imgs = sobel_process(orig_imgs, config.include_rgb, using_IR=config.using_IR) else: imgs = orig_imgs with torch.no_grad(): x_outs_all = net(imgs) x_outs = x_outs_all[head_i] x_outs = x_outs.cpu().numpy() flat_preds = np.argmax(x_outs, axis=1) n, h, w = flat_preds.shape num_imgs_curr = flat_preds.shape[0] reordered_preds = np.zeros((num_imgs_curr, h, w), dtype=flat_targets.dtype)
actual_batch_start = d_i * curr_batch_sz actual_batch_end = actual_batch_start + curr_batch_sz all_imgs[actual_batch_start:actual_batch_end, :, :, :] = \ imgs_curr.cuda() all_imgs_tf[actual_batch_start:actual_batch_end, :, :, :] = \ imgs_tf_curr.cuda() if not (curr_batch_sz == config.dataloader_batch_sz): print("last batch sz %d" % curr_batch_sz) curr_total_batch_sz = curr_batch_sz * config.num_dataloaders all_imgs = all_imgs[:curr_total_batch_sz, :, :, :] all_imgs_tf = all_imgs_tf[:curr_total_batch_sz, :, :, :] all_imgs = sobel_process(all_imgs, config.include_rgb) all_imgs_tf = sobel_process(all_imgs_tf, config.include_rgb) x_outs = net(all_imgs, head=head) x_tf_outs = net(all_imgs_tf, head=head) avg_loss_batch = None # avg over the sub_heads avg_loss_no_lamb_batch = None for i in xrange(config.num_sub_heads): loss, loss_no_lamb = IID_loss(x_outs[i], x_tf_outs[i], lamb=config.lamb) if avg_loss_batch is None: avg_loss_batch = loss avg_loss_no_lamb_batch = loss_no_lamb else: avg_loss_batch += loss
all_mask_img1[ actual_batch_start:actual_batch_end, :, :] = mask_img1 if not (curr_batch_sz == config.dataloader_batch_sz) and (e_i == next_epoch): print("last batch sz %d" % curr_batch_sz) curr_total_batch_sz = curr_batch_sz * config.num_dataloaders all_img1 = all_img1[:curr_total_batch_sz, :, :, :] all_img2 = all_img2[:curr_total_batch_sz, :, :, :] all_affine2_to_1 = all_affine2_to_1[:curr_total_batch_sz, :, :] all_mask_img1 = all_mask_img1[:curr_total_batch_sz, :, :] if (not config.no_sobel): all_img1 = sobel_process(all_img1, config.include_rgb, using_IR=config.using_IR) all_img2 = sobel_process(all_img2, config.include_rgb, using_IR=config.using_IR) x1_outs = net(all_img1) x2_outs = net(all_img2) avg_loss_batch = None # avg over the heads avg_loss_no_lamb_batch = None for i in xrange(config.num_sub_heads): loss, loss_no_lamb = loss_fn( x1_outs[i], x2_outs[i], all_affine2_to_1=all_affine2_to_1,
def train(): dataloaders_head_A, mapping_assignment_dataloader, mapping_test_dataloader = \ segmentation_create_dataloaders(config) dataloaders_head_B = dataloaders_head_A # unlike for clustering datasets net = archs.__dict__[config.arch](config) if config.restart: dict = torch.load(os.path.join(config.out_dir, dict_name), map_location=lambda storage, loc: storage) net.load_state_dict(dict["net"]) net.cuda() net = torch.nn.DataParallel(net) net.train() optimiser = get_opt(config.opt)(net.module.parameters(), lr=config.lr) if config.restart: optimiser.load_state_dict(dict["optimiser"]) heads = ["A", "B"] if hasattr(config, "head_B_first") and config.head_B_first: heads = ["B", "A"] # Results # ---------------------------------------------------------------------- if config.restart: next_epoch = config.last_epoch + 1 print("starting from epoch %d" % next_epoch) config.epoch_acc = config.epoch_acc[:next_epoch] # in case we overshot config.epoch_avg_subhead_acc = config.epoch_avg_subhead_acc[: next_epoch] config.epoch_stats = config.epoch_stats[:next_epoch] config.epoch_loss_head_A = config.epoch_loss_head_A[:(next_epoch - 1)] config.epoch_loss_no_lamb_head_A = config.epoch_loss_no_lamb_head_A[:( next_epoch - 1)] config.epoch_loss_head_B = config.epoch_loss_head_B[:(next_epoch - 1)] config.epoch_loss_no_lamb_head_B = config.epoch_loss_no_lamb_head_B[:( next_epoch - 1)] else: config.epoch_acc = [] config.epoch_avg_subhead_acc = [] config.epoch_stats = [] config.epoch_loss_head_A = [] config.epoch_loss_no_lamb_head_A = [] config.epoch_loss_head_B = [] config.epoch_loss_no_lamb_head_B = [] _ = segmentation_eval( config, net, mapping_assignment_dataloader=mapping_assignment_dataloader, mapping_test_dataloader=mapping_test_dataloader, sobel=(not config.no_sobel), using_IR=config.using_IR) print("Pre: time %s: \n %s" % (datetime.now(), nice(config.epoch_stats[-1]))) sys.stdout.flush() next_epoch = 1 fig, axarr = plt.subplots(6, sharex=False, figsize=(20, 20)) if not config.use_uncollapsed_loss: print("using condensed loss (default)") loss_fn = IID_segmentation_loss else: print("using uncollapsed loss!") loss_fn = IID_segmentation_loss_uncollapsed # Train # ------------------------------------------------------------------------ for e_i in xrange(next_epoch, config.num_epochs): print("Starting e_i: %d %s" % (e_i, datetime.now())) sys.stdout.flush() if e_i in config.lr_schedule: optimiser = update_lr(optimiser, lr_mult=config.lr_mult) for head_i in range(2): head = heads[head_i] if head == "A": dataloaders = dataloaders_head_A epoch_loss = config.epoch_loss_head_A epoch_loss_no_lamb = config.epoch_loss_no_lamb_head_A lamb = config.lamb_A elif head == "B": dataloaders = dataloaders_head_B epoch_loss = config.epoch_loss_head_B epoch_loss_no_lamb = config.epoch_loss_no_lamb_head_B lamb = config.lamb_B iterators = (d for d in dataloaders) b_i = 0 avg_loss = 0. # over heads and head_epochs (and sub_heads) avg_loss_no_lamb = 0. avg_loss_count = 0 for tup in itertools.izip(*iterators): net.module.zero_grad() if not config.no_sobel: pre_channels = config.in_channels - 1 else: pre_channels = config.in_channels all_img1 = torch.zeros(config.batch_sz, pre_channels, config.input_sz, config.input_sz).to( torch.float32).cuda() all_img2 = torch.zeros(config.batch_sz, pre_channels, config.input_sz, config.input_sz).to( torch.float32).cuda() all_affine2_to_1 = torch.zeros(config.batch_sz, 2, 3).to(torch.float32).cuda() all_mask_img1 = torch.zeros(config.batch_sz, config.input_sz, config.input_sz).to( torch.float32).cuda() curr_batch_sz = tup[0][0].shape[0] for d_i in xrange(config.num_dataloaders): img1, img2, affine2_to_1, mask_img1 = tup[d_i] assert (img1.shape[0] == curr_batch_sz) actual_batch_start = d_i * curr_batch_sz actual_batch_end = actual_batch_start + curr_batch_sz all_img1[ actual_batch_start:actual_batch_end, :, :, :] = img1 all_img2[ actual_batch_start:actual_batch_end, :, :, :] = img2 all_affine2_to_1[actual_batch_start: actual_batch_end, :, :] = affine2_to_1 all_mask_img1[ actual_batch_start:actual_batch_end, :, :] = mask_img1 if not (curr_batch_sz == config.dataloader_batch_sz) and (e_i == next_epoch): print("last batch sz %d" % curr_batch_sz) curr_total_batch_sz = curr_batch_sz * config.num_dataloaders # times 2 all_img1 = all_img1[:curr_total_batch_sz, :, :, :] all_img2 = all_img2[:curr_total_batch_sz, :, :, :] all_affine2_to_1 = all_affine2_to_1[:curr_total_batch_sz, :, :] all_mask_img1 = all_mask_img1[:curr_total_batch_sz, :, :] if (not config.no_sobel): all_img1 = sobel_process(all_img1, config.include_rgb, using_IR=config.using_IR) all_img2 = sobel_process(all_img2, config.include_rgb, using_IR=config.using_IR) x1_outs = net(all_img1, head=head) x2_outs = net(all_img2, head=head) avg_loss_batch = None # avg over the heads avg_loss_no_lamb_batch = None for i in xrange(config.num_sub_heads): loss, loss_no_lamb = loss_fn( x1_outs[i], x2_outs[i], all_affine2_to_1=all_affine2_to_1, all_mask_img1=all_mask_img1, lamb=lamb, half_T_side_dense=config.half_T_side_dense, half_T_side_sparse_min=config.half_T_side_sparse_min, half_T_side_sparse_max=config.half_T_side_sparse_max) if avg_loss_batch is None: avg_loss_batch = loss avg_loss_no_lamb_batch = loss_no_lamb else: avg_loss_batch += loss avg_loss_no_lamb_batch += loss_no_lamb avg_loss_batch /= config.num_sub_heads avg_loss_no_lamb_batch /= config.num_sub_heads if ((b_i % 100) == 0) or (e_i == next_epoch): print( "Model ind %d epoch %d head %s batch: %d avg loss %f avg loss no " "lamb %f " "time %s" % \ (config.model_ind, e_i, head, b_i, avg_loss_batch.item(), avg_loss_no_lamb_batch.item(), datetime.now())) sys.stdout.flush() if not np.isfinite(avg_loss_batch.item()): print("Loss is not finite... %s:" % str(avg_loss_batch)) exit(1) avg_loss += avg_loss_batch.item() avg_loss_no_lamb += avg_loss_no_lamb_batch.item() avg_loss_count += 1 avg_loss_batch.backward() optimiser.step() torch.cuda.empty_cache() b_i += 1 if b_i == 2 and config.test_code: break avg_loss = float(avg_loss / avg_loss_count) avg_loss_no_lamb = float(avg_loss_no_lamb / avg_loss_count) epoch_loss.append(avg_loss) epoch_loss_no_lamb.append(avg_loss_no_lamb) # Eval # ----------------------------------------------------------------------- is_best = segmentation_eval( config, net, mapping_assignment_dataloader=mapping_assignment_dataloader, mapping_test_dataloader=mapping_test_dataloader, sobel=(not config.no_sobel), using_IR=config.using_IR) print("Pre: time %s: \n %s" % (datetime.now(), nice(config.epoch_stats[-1]))) sys.stdout.flush() axarr[0].clear() axarr[0].plot(config.epoch_acc) axarr[0].set_title("acc (best), top: %f" % max(config.epoch_acc)) axarr[1].clear() axarr[1].plot(config.epoch_avg_subhead_acc) axarr[1].set_title("acc (avg), top: %f" % max(config.epoch_avg_subhead_acc)) axarr[2].clear() axarr[2].plot(config.epoch_loss_head_A) axarr[2].set_title("Loss head A") axarr[3].clear() axarr[3].plot(config.epoch_loss_no_lamb_head_A) axarr[3].set_title("Loss no lamb head A") axarr[4].clear() axarr[4].plot(config.epoch_loss_head_B) axarr[4].set_title("Loss head B") axarr[5].clear() axarr[5].plot(config.epoch_loss_no_lamb_head_B) axarr[5].set_title("Loss no lamb head B") fig.canvas.draw_idle() fig.savefig(os.path.join(config.out_dir, "plots.png")) if is_best or (e_i % config.save_freq == 0): net.module.cpu() save_dict = { "net": net.module.state_dict(), "optimiser": optimiser.state_dict() } if e_i % config.save_freq == 0: torch.save(save_dict, os.path.join(config.out_dir, "latest.pytorch")) config.last_epoch = e_i # for last saved version if is_best: torch.save(save_dict, os.path.join(config.out_dir, "best.pytorch")) with open(os.path.join(config.out_dir, "best_config.pickle"), 'wb') as outfile: pickle.dump(config, outfile) with open(os.path.join(config.out_dir, "best_config.txt"), "w") as text_file: text_file.write("%s" % config) net.module.cuda() with open(os.path.join(config.out_dir, "config.pickle"), 'wb') as outfile: pickle.dump(config, outfile) with open(os.path.join(config.out_dir, "config.txt"), "w") as text_file: text_file.write("%s" % config) if config.test_code: exit(0)
if e_i in config.lr_schedule: optimiser = update_lr(optimiser, lr_mult=config.lr_mult) avg_loss = 0. # over heads and head_epochs (and sub_heads) avg_loss_count = 0 sys.stdout.flush() iterators = (d for d in train_dataloaders) b_i = 0 for tup in itertools.izip(*iterators): net.module.zero_grad() # no sobel yet imgs_orig = sobel_process(tup[0][0].cuda(), config.include_rgb) imgs_pos = sobel_process(tup[1][0].cuda(), config.include_rgb) imgs_neg = sobel_process(tup[2][0].cuda(), config.include_rgb) outs_orig = net(imgs_orig) outs_pos = net(imgs_pos) outs_neg = net(imgs_neg) curr_loss = triplets_loss(outs_orig, outs_pos, outs_neg) if ((b_i % 100) == 0) or (e_i == next_epoch and b_i < 10): print("Model ind %d epoch %d batch %d " "loss %f time %s" % \ (config.model_ind, e_i, b_i, curr_loss.item(), datetime.now())) sys.stdout.flush()
def train_kmeans(config, net, test_dataloader): num_imgs = len(test_dataloader.dataset) max_num_pixels_per_img = int(config.max_num_kmeans_samples / num_imgs) features_all = np.zeros( (config.max_num_kmeans_samples, net.module.features_sz), dtype=np.float32) actual_num_features = 0 # discard the label information in the dataloader for i, tup in enumerate(test_dataloader): if (config.verbose and i < 10) or (i % int(len(test_dataloader) / 10) == 0): print("(kmeans_segmentation_eval) batch %d time %s" % (i, datetime.now())) sysout.flush() imgs, _, mask = tup # test dataloader, cpu tensors imgs = imgs.cuda() mask = mask.numpy().astype(np.bool) # mask = mask.numpy().astype(np.bool) num_unmasked = mask.sum() if not config.no_sobel: imgs = sobel_process(imgs, config.include_rgb, using_IR=config.using_IR) # now rgb(ir) and/or sobel with torch.no_grad(): # penultimate = features x_out = net(imgs, penultimate=True).cpu().numpy() if config.verbose and i < 2: print("(kmeans_segmentation_eval) through model %d time %s" % (i, datetime.now())) sysout.flush() num_imgs_batch = x_out.shape[0] x_out = x_out.transpose((0, 2, 3, 1)) # features last x_out = x_out[mask, :] if config.verbose and i < 2: print("(kmeans_segmentation_eval) applied mask %d time %s" % (i, datetime.now())) sysout.flush() if i == 0: assert (x_out.shape[1] == net.module.features_sz) assert (x_out.shape[0] == num_unmasked) # select pixels randomly, and record how many selected num_selected = min(num_unmasked, num_imgs_batch * max_num_pixels_per_img) selected = np.random.choice(num_selected, replace=False) x_out = x_out[selected, :] if config.verbose and i < 2: print("(kmeans_segmentation_eval) applied select %d time %s" % (i, datetime.now())) sysout.flush() features_all[actual_num_features:actual_num_features + num_selected, :] = \ x_out actual_num_features += num_selected if config.verbose and i < 2: print("(kmeans_segmentation_eval) stored %d time %s" % (i, datetime.now())) sysout.flush() assert (actual_num_features <= config.max_num_kmeans_samples) features_all = features_all[:actual_num_features, :] if config.verbose: print("running kmeans") sysout.flush() kmeans = MiniBatchKMeans(n_clusters=config.gt_k, verbose=config.verbose).fit(features_all) return kmeans
def apply_trained_kmeans(config, net, test_dataloader, kmeans): if config.verbose: print("starting inference") sysout.flush() # on the entire test dataset num_imgs = len(test_dataloader.dataset) max_num_samples = num_imgs * config.input_sz * config.input_sz preds_all = torch.zeros(max_num_samples, dtype=torch.int32).cuda() targets_all = torch.zeros(max_num_samples, dtype=torch.int32).cuda() actual_num_unmasked = 0 # discard the label information in the dataloader for i, tup in enumerate(test_dataloader): if (config.verbose and i < 10) or (i % int(len(test_dataloader) / 10) == 0): print("(apply_trained_kmeans) batch %d time %s" % (i, datetime.now())) sysout.flush() imgs, targets, mask = tup # test dataloader, cpu tensors imgs, mask_cuda, targets, mask_np = imgs.cuda(), mask.cuda(), \ targets.cuda(), mask.numpy().astype( np.bool) num_unmasked = mask_cuda.sum().item() if not config.no_sobel: imgs = sobel_process(imgs, config.include_rgb, using_IR=config.using_IR) # now rgb(ir) and/or sobel with torch.no_grad(): # penultimate = features x_out = net(imgs, penultimate=True).cpu().numpy() x_out = x_out.transpose((0, 2, 3, 1)) # features last x_out = x_out[mask_np, :] targets = targets.masked_select(mask_cuda) # can do because flat assert (x_out.shape == (num_unmasked, net.module.features_sz)) preds = torch.from_numpy(kmeans.predict(x_out)).cuda() preds_all[actual_num_unmasked:actual_num_unmasked + num_unmasked] = preds targets_all[actual_num_unmasked:actual_num_unmasked + num_unmasked] = targets actual_num_unmasked += num_unmasked preds_all = preds_all[:actual_num_unmasked] targets_all = targets_all[:actual_num_unmasked] torch.cuda.empty_cache() # permutation, not many-to-one match = _hungarian_match(preds_all, targets_all, preds_k=config.gt_k, targets_k=config.gt_k) torch.cuda.empty_cache() # do in cpu because of RAM reordered_preds = torch.zeros(actual_num_unmasked, dtype=preds_all.dtype) for pred_i, target_i in match: selected = (preds_all == pred_i).cpu() reordered_preds[selected] = target_i reordered_preds = reordered_preds.cuda() # this checks values acc = _acc(reordered_preds, targets_all, config.gt_k, verbose=config.verbose) if GET_NMI_ARI: nmi, ari = _nmi(reordered_preds, targets_all), \ _ari(reordered_preds, targets_all) else: nmi, ari = -1., -1. reordered_masses = np.zeros(config.gt_k) for c in xrange(config.gt_k): reordered_masses[c] = float( (reordered_preds == c).sum()) / actual_num_unmasked return acc, nmi, ari, reordered_masses
is_best = False if e_i in config.lr_schedule: optimiser = update_lr(optimiser, lr_mult=config.lr_mult) avg_loss = 0. # over epoch for b_i, tup in enumerate(dataloader): net.module.zero_grad() img, mask = tup # cuda # no need for requires_grad or Variable (torch 0.4.1) if (not config.no_sobel): img = sobel_process(img, config.include_rgb, using_IR=config.using_IR) centre, other, position_gt = doersch_set_patches( input_sz=config.input_sz, patch_side=config.doersch_patch_side) position_pred = net(img, centre=centre, other=other) loss = doersch_loss(position_pred, centre, other, position_gt, mask, crossent=crossent, verbose=config.verbose) if ((b_i % 100) == 0) or (e_i == next_epoch):
if not os.path.exists(render_out_dir): os.makedirs(render_out_dir) results_f = os.path.join(render_out_dir, "results.txt") iterators = (d for d in [dataloader, render_dataloader]) for tup in itertools.izip(*iterators): train_batch = tup[0] render_batch = tup[1] imgs = train_batch[0].cuda() orig_imgs = render_batch[0] if sobel: imgs = sobel_process(imgs, config.include_rgb, using_IR=using_IR) flat_targets = train_batch[1] with torch.no_grad(): x_outs = net(imgs) assert (x_outs[0].shape[1] == config.output_k) assert (len(x_outs[0].shape) == 2) x_outs_curr = x_outs[best_head] flat_preds_curr = torch.argmax(x_outs_curr, dim=1) # along output_k with open(results_f, "w") as f: for i, img_i in enumerate(img_inds): img = orig_imgs[img_i].numpy()
def main(): parser = argparse.ArgumentParser() parser.add_argument("--model_ind", type=int, required=True) parser.add_argument("--arch", type=str, required=True) parser.add_argument("--head_lr", type=float, required=True) parser.add_argument("--trunk_lr", type=float, required=True) parser.add_argument("--num_epochs", type=int, default=3200) parser.add_argument("--new_batch_sz", type=int, default=-1) parser.add_argument("--old_model_ind", type=int, required=True) parser.add_argument("--penultimate_features", default=False, action="store_true") parser.add_argument("--random_affine", default=False, action="store_true") parser.add_argument("--affine_p", type=float, default=0.5) parser.add_argument("--cutout", default=False, action="store_true") parser.add_argument("--cutout_p", type=float, default=0.5) parser.add_argument("--cutout_max_box", type=float, default=0.5) parser.add_argument("--restart", default=False, action="store_true") parser.add_argument("--lr_schedule", type=int, nargs="+", default=[]) parser.add_argument("--lr_mult", type=float, default=0.5) parser.add_argument("--restart_new_model_ind", default=False, action="store_true") parser.add_argument("--new_model_ind", type=int, default=0) parser.add_argument("--out_root", type=str, default="/scratch/shared/slow/xuji/iid_private") config = parser.parse_args() # new config # Setup ---------------------------------------------------------------------- config.contiguous_sz = 10 # Tencrop config.out_dir = os.path.join(config.out_root, str(config.model_ind)) if not os.path.exists(config.out_dir): os.makedirs(config.out_dir) if config.restart: given_config = config reloaded_config_path = os.path.join(given_config.out_dir, "config.pickle") print("Loading restarting config from: %s" % reloaded_config_path) with open(reloaded_config_path, "rb") as config_f: config = pickle.load(config_f) assert (config.model_ind == given_config.model_ind) config.restart = True config.num_epochs = given_config.num_epochs # train for longer config.restart_new_model_ind = given_config.restart_new_model_ind config.new_model_ind = given_config.new_model_ind start_epoch = config.last_epoch + 1 print("...restarting from epoch %d" % start_epoch) # in case we overshot without saving config.epoch_acc = config.epoch_acc[:start_epoch] config.epoch_loss = config.epoch_loss[:start_epoch] else: config.epoch_acc = [] config.epoch_loss = [] start_epoch = 0 # old config only used retrospectively for setting up model at start reloaded_config_path = os.path.join( os.path.join(config.out_root, str(config.old_model_ind)), "config.pickle") print("Loading old features config from: %s" % reloaded_config_path) with open(reloaded_config_path, "rb") as config_f: old_config = pickle.load(config_f) assert (old_config.model_ind == config.old_model_ind) if config.new_batch_sz == -1: config.new_batch_sz = old_config.batch_sz fig, axarr = plt.subplots(2, sharex=False, figsize=(20, 20)) # Data ----------------------------------------------------------------------- assert (old_config.dataset == "STL10") # make supervised data: train on train, test on test, unlabelled is unused tf1, tf2, tf3 = sobel_make_transforms(old_config, random_affine=config.random_affine, cutout=config.cutout, cutout_p=config.cutout_p, cutout_max_box=config.cutout_max_box, affine_p=config.affine_p) dataset_class = torchvision.datasets.STL10 train_data = dataset_class( root=old_config.dataset_root, transform=tf2, # also could use tf1 split="train") train_loader = torch.utils.data.DataLoader(train_data, batch_size=config.new_batch_sz, shuffle=True, num_workers=0, drop_last=False) test_data = dataset_class(root=old_config.dataset_root, transform=None, split="test") test_data = TenCropAndFinish(test_data, input_sz=old_config.input_sz, include_rgb=old_config.include_rgb) test_loader = torch.utils.data.DataLoader( test_data, batch_size=config.new_batch_sz, # full batch shuffle=False, num_workers=0, drop_last=False) # Model ---------------------------------------------------------------------- net_features = archs.__dict__[old_config.arch](old_config) if not config.restart: model_path = os.path.join(old_config.out_dir, "best_net.pytorch") net_features.load_state_dict( torch.load(model_path, map_location=lambda storage, loc: storage)) dlen = get_dlen(net_features, train_loader, include_rgb=old_config.include_rgb, penultimate_features=config.penultimate_features) print("dlen: %d" % dlen) assert (config.arch == "SupHead5") net = SupHead5(net_features, dlen=dlen, gt_k=old_config.gt_k) if config.restart: print("restarting from latest net") model_path = os.path.join(config.out_dir, "latest_net.pytorch") net.load_state_dict( torch.load(model_path, map_location=lambda storage, loc: storage)) net.cuda() net = torch.nn.DataParallel(net) opt_trunk = torch.optim.Adam(net.module.trunk.parameters(), lr=config.trunk_lr) opt_head = torch.optim.Adam(net.module.head.parameters(), lr=(config.head_lr)) if config.restart: print("restarting from latest optimiser") optimiser_states = torch.load( os.path.join(config.out_dir, "latest_optimiser.pytorch")) opt_trunk.load_state_dict(optimiser_states["opt_trunk"]) opt_head.load_state_dict(optimiser_states["opt_head"]) else: print("using new optimiser state") criterion = nn.CrossEntropyLoss().cuda() if not config.restart: net.eval() acc = assess_acc_block( net, test_loader, gt_k=old_config.gt_k, include_rgb=old_config.include_rgb, penultimate_features=config.penultimate_features, contiguous_sz=config.contiguous_sz) print("pre: model %d old model %d, acc %f time %s" % (config.model_ind, config.old_model_ind, acc, datetime.now())) sys.stdout.flush() config.epoch_acc.append(acc) if config.restart_new_model_ind: assert (config.restart) config.model_ind = config.new_model_ind # old_model_ind stays same config.out_dir = os.path.join(config.out_root, str(config.model_ind)) print("restarting as model %d" % config.model_ind) if not os.path.exists(config.out_dir): os.makedirs(config.out_dir) # Train ---------------------------------------------------------------------- for e_i in xrange(start_epoch, config.num_epochs): net.train() if e_i in config.lr_schedule: print("e_i %d, multiplying lr for opt trunk and head by %f" % (e_i, config.lr_mult)) opt_trunk = update_lr(opt_trunk, lr_mult=config.lr_mult) opt_head = update_lr(opt_head, lr_mult=config.lr_mult) if not hasattr(config, "lr_changes"): config.lr_changes = [] config.lr_changes.append((e_i, config.lr_mult)) avg_loss = 0. num_batches = len(train_loader) for i, (imgs, targets) in enumerate(train_loader): imgs = sobel_process(imgs.cuda(), old_config.include_rgb) targets = targets.cuda() x_out = net(imgs, penultimate_features=config.penultimate_features) loss = criterion(x_out, targets) avg_loss += float(loss.data) opt_trunk.zero_grad() opt_head.zero_grad() loss.backward() opt_trunk.step() opt_head.step() if (i % 100 == 0) or (e_i == start_epoch): print("batch %d of %d, loss %f, time %s" % (i, num_batches, float(loss.data), datetime.now())) sys.stdout.flush() avg_loss /= num_batches net.eval() acc = assess_acc_block( net, test_loader, gt_k=old_config.gt_k, include_rgb=old_config.include_rgb, penultimate_features=config.penultimate_features, contiguous_sz=config.contiguous_sz) print( "model %d old model %d epoch %d acc %f time %s" % (config.model_ind, config.old_model_ind, e_i, acc, datetime.now())) sys.stdout.flush() is_best = False if acc > max(config.epoch_acc): is_best = True config.epoch_acc.append(acc) config.epoch_loss.append(avg_loss) axarr[0].clear() axarr[0].plot(config.epoch_acc) axarr[0].set_title("Acc") axarr[1].clear() axarr[1].plot(config.epoch_loss) axarr[1].set_title("Loss") fig.canvas.draw_idle() fig.savefig(os.path.join(config.out_dir, "plots.png")) if is_best or (e_i % 10 == 0): net.module.cpu() if is_best: torch.save(net.module.state_dict(), os.path.join(config.out_dir, "best_net.pytorch")) torch.save( { "opt_head": opt_head.state_dict(), "opt_trunk": opt_trunk.state_dict() }, os.path.join(config.out_dir, "best_optimiser.pytorch")) # save model sparingly for this script if e_i % 10 == 0: torch.save(net.module.state_dict(), os.path.join(config.out_dir, "latest_net.pytorch")) torch.save( { "opt_head": opt_head.state_dict(), "opt_trunk": opt_trunk.state_dict() }, os.path.join(config.out_dir, "latest_optimiser.pytorch")) net.module.cuda() config.last_epoch = e_i # for last saved version with open(os.path.join(config.out_dir, "config.pickle"), 'w') as outfile: pickle.dump(config, outfile) with open(os.path.join(config.out_dir, "config.txt"), "w") as text_file: text_file.write("%s" % config)