def debug(self, batch, output, iter_id, dataset):
opt = self.opt
if 'pre_hm' in batch:
output.update({'pre_hm': batch['pre_hm']})
dets = generic_decode(output, K=opt.K, opt=opt)
for k in dets:
dets[k] = dets[k].detach().cpu().numpy()
dets_gt = batch['meta']['gt_det']
for i in range(1):
debugger = Debugger(opt=opt, dataset=dataset)
img = batch['image'][i].detach().cpu().numpy().transpose(1, 2, 0)
img = np.clip(((img * dataset.std + dataset.mean) * 255.), 0,
255).astype(np.uint8)
pred = debugger.gen_colormap(
output['hm'][i].detach().cpu().numpy())
gt = debugger.gen_colormap(batch['hm'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'pred_hm')
debugger.add_blend_img(img, gt, 'gt_hm')
if 'pre_img' in batch:
pre_img = batch['pre_img'][i].detach().cpu().numpy().transpose(
1, 2, 0)
pre_img = np.clip(
((pre_img * dataset.std + dataset.mean) * 255), 0,
255).astype(np.uint8)
debugger.add_img(pre_img, 'pre_img_pred')
debugger.add_img(pre_img, 'pre_img_gt')
if 'pre_hm' in batch:
pre_hm = debugger.gen_colormap(
batch['pre_hm'][i].detach().cpu().numpy())
debugger.add_blend_img(pre_img, pre_hm, 'pre_hm')
debugger.add_img(img, img_id='out_pred')
if 'ltrb_amodal' in opt.heads:
debugger.add_img(img, img_id='out_pred_amodal')
debugger.add_img(img, img_id='out_gt_amodal')
# Predictions
for k in range(len(dets['scores'][i])):
if dets['scores'][i, k] > opt.vis_thresh:
debugger.add_coco_bbox(dets['bboxes'][i, k] *
opt.down_ratio,
dets['clses'][i, k],
dets['scores'][i, k],
img_id='out_pred')
if 'ltrb_amodal' in opt.heads:
debugger.add_coco_bbox(dets['bboxes_amodal'][i, k] *
opt.down_ratio,
dets['clses'][i, k],
dets['scores'][i, k],
img_id='out_pred_amodal')
if 'hps' in opt.heads and int(dets['clses'][i, k]) == 0:
debugger.add_coco_hp(dets['hps'][i, k] *
opt.down_ratio,
img_id='out_pred')
if 'tracking' in opt.heads:
debugger.add_arrow(dets['cts'][i][k] * opt.down_ratio,
dets['tracking'][i][k] *
opt.down_ratio,
img_id='out_pred')
debugger.add_arrow(dets['cts'][i][k] * opt.down_ratio,
dets['tracking'][i][k] *
opt.down_ratio,
img_id='pre_img_pred')
# Ground truth
debugger.add_img(img, img_id='out_gt')
for k in range(len(dets_gt['scores'][i])):
if dets_gt['scores'][i][k] > opt.vis_thresh:
debugger.add_coco_bbox(dets_gt['bboxes'][i][k] *
opt.down_ratio,
dets_gt['clses'][i][k],
dets_gt['scores'][i][k],
img_id='out_gt')
if 'ltrb_amodal' in opt.heads:
debugger.add_coco_bbox(dets_gt['bboxes_amodal'][i, k] *
opt.down_ratio,
dets_gt['clses'][i, k],
dets_gt['scores'][i, k],
img_id='out_gt_amodal')
if 'hps' in opt.heads and \
(int(dets['clses'][i, k]) == 0):
debugger.add_coco_hp(dets_gt['hps'][i][k] *
opt.down_ratio,
img_id='out_gt')
if 'tracking' in opt.heads:
debugger.add_arrow(
dets_gt['cts'][i][k] * opt.down_ratio,
dets_gt['tracking'][i][k] * opt.down_ratio,
img_id='out_gt')
debugger.add_arrow(
dets_gt['cts'][i][k] * opt.down_ratio,
dets_gt['tracking'][i][k] * opt.down_ratio,
img_id='pre_img_gt')
if 'hm_hp' in opt.heads:
pred = debugger.gen_colormap_hp(
output['hm_hp'][i].detach().cpu().numpy())
gt = debugger.gen_colormap_hp(
batch['hm_hp'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'pred_hmhp')
debugger.add_blend_img(img, gt, 'gt_hmhp')
if 'rot' in opt.heads and 'dim' in opt.heads and 'dep' in opt.heads:
dets_gt = {k: dets_gt[k].cpu().numpy() for k in dets_gt}
calib = batch['meta']['calib'].detach().numpy() \
if 'calib' in batch['meta'] else None
det_pred = generic_post_process(
opt, dets, batch['meta']['c'].cpu().numpy(),
batch['meta']['s'].cpu().numpy(), output['hm'].shape[2],
output['hm'].shape[3], self.opt.num_classes, calib)
det_gt = generic_post_process(opt, dets_gt,
batch['meta']['c'].cpu().numpy(),
batch['meta']['s'].cpu().numpy(),
output['hm'].shape[2],
output['hm'].shape[3],
self.opt.num_classes, calib)
debugger.add_3d_detection(batch['meta']['img_path'][i],
batch['meta']['flipped'][i],
det_pred[i],
calib[i],
vis_thresh=opt.vis_thresh,
img_id='add_pred')
debugger.add_3d_detection(batch['meta']['img_path'][i],
batch['meta']['flipped'][i],
det_gt[i],
calib[i],
vis_thresh=opt.vis_thresh,
img_id='add_gt')
debugger.add_bird_views(det_pred[i],
det_gt[i],
vis_thresh=opt.vis_thresh,
img_id='bird_pred_gt')
if opt.debug == 4:
debugger.save_all_imgs(opt.debug_dir,
prefix='{}'.format(iter_id))
else:
debugger.show_all_imgs(pause=True)
def debug(self, batch, output, iter_id):
opt = self.opt
reg = output['reg'] if opt.reg_offset else None
hm_hp = output['hm_hp'] if opt.hm_hp else None
hp_offset = output['hp_offset'] if opt.reg_hp_offset else None
dets = multi_pose_decode(output['hm'],
output['wh'],
output['hps'],
reg=reg,
hm_hp=hm_hp,
hp_offset=hp_offset,
K=opt.K)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets[:, :, :4] *= opt.input_res / opt.output_res
dets[:, :, 5:39] *= opt.input_res / opt.output_res
dets_gt = batch['meta']['gt_det'].numpy().reshape(1, -1, dets.shape[2])
dets_gt[:, :, :4] *= opt.input_res / opt.output_res
dets_gt[:, :, 5:39] *= opt.input_res / opt.output_res
for i in range(1):
debugger = Debugger(dataset=opt.dataset,
ipynb=(opt.debug == 3),
theme=opt.debugger_theme)
img = batch['input'][i].detach().cpu().numpy().transpose(1, 2, 0)
img = np.clip(((img * opt.std + opt.mean) * 255.), 0,
255).astype(np.uint8)
pred = debugger.gen_colormap(
output['hm'][i].detach().cpu().numpy())
gt = debugger.gen_colormap(batch['hm'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'pred_hm')
debugger.add_blend_img(img, gt, 'gt_hm')
debugger.add_img(img, img_id='out_pred')
for k in range(len(dets[i])):
if dets[i, k, 4] > opt.center_thresh:
debugger.add_coco_bbox(dets[i, k, :4],
dets[i, k, -1],
dets[i, k, 4],
img_id='out_pred')
debugger.add_coco_hp(dets[i, k, 5:39], img_id='out_pred')
debugger.add_img(img, img_id='out_gt')
for k in range(len(dets_gt[i])):
if dets_gt[i, k, 4] > opt.center_thresh:
debugger.add_coco_bbox(dets_gt[i, k, :4],
dets_gt[i, k, -1],
dets_gt[i, k, 4],
img_id='out_gt')
debugger.add_coco_hp(dets_gt[i, k, 5:39], img_id='out_gt')
if opt.hm_hp:
pred = debugger.gen_colormap_hp(
output['hm_hp'][i].detach().cpu().numpy())
gt = debugger.gen_colormap_hp(
batch['hm_hp'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'pred_hmhp')
debugger.add_blend_img(img, gt, 'gt_hmhp')
if opt.debug == 4:
debugger.save_all_imgs(opt.debug_dir,
prefix='{}'.format(iter_id))
else:
debugger.show_all_imgs(pause=True)
def debug(self, batch, output, iter_id):
opt = self.opt
wh = output['wh'] if opt.reg_bbox else None
reg = output['reg'] if opt.reg_offset else None
dets = ddd_decode(output['hm'],
output['rot'],
output['dep'],
output['dim'],
wh=wh,
reg=reg,
K=opt.K)
# x, y, score, r1-r8, depth, dim1-dim3, cls
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
calib = batch['meta']['calib'].detach().numpy()
# x, y, score, rot, depth, dim1, dim2, dim3
# if opt.dataset == 'gta':
# dets[:, 12:15] /= 3
dets_pred = ddd_post_process(dets.copy(),
batch['meta']['c'].detach().numpy(),
batch['meta']['s'].detach().numpy(),
calib, opt)
dets_gt = ddd_post_process(
batch['meta']['gt_det'].detach().numpy().copy(),
batch['meta']['c'].detach().numpy(),
batch['meta']['s'].detach().numpy(), calib, opt)
#for i in range(input.size(0)):
for i in range(1):
debugger = Debugger(dataset=opt.dataset,
ipynb=(opt.debug == 3),
theme=opt.debugger_theme)
img = batch['input'][i].detach().cpu().numpy().transpose(1, 2, 0)
img = ((img * self.opt.std + self.opt.mean) * 255.).astype(
np.uint8)
pred = debugger.gen_colormap(
output['hm'][i].detach().cpu().numpy())
gt = debugger.gen_colormap(batch['hm'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'hm_pred')
debugger.add_blend_img(img, gt, 'hm_gt')
# decode
debugger.add_ct_detection(img,
dets[i],
show_box=opt.reg_bbox,
center_thresh=opt.center_thresh,
img_id='det_pred')
debugger.add_ct_detection(
img,
batch['meta']['gt_det'][i].cpu().numpy().copy(),
show_box=opt.reg_bbox,
img_id='det_gt')
debugger.add_3d_detection(batch['meta']['image_path'][i],
dets_pred[i],
calib[i],
center_thresh=opt.center_thresh,
img_id='add_pred')
debugger.add_3d_detection(batch['meta']['image_path'][i],
dets_gt[i],
calib[i],
center_thresh=opt.center_thresh,
img_id='add_gt')
# debugger.add_bird_view(
# dets_pred[i], center_thresh=opt.center_thresh, img_id='bird_pred')
# debugger.add_bird_view(dets_gt[i], img_id='bird_gt')
debugger.add_bird_views(dets_pred[i],
dets_gt[i],
center_thresh=opt.center_thresh,
img_id='bird_pred_gt')
# debugger.add_blend_img(img, pred, 'out', white=True)
debugger.compose_vis_add(batch['meta']['image_path'][i],
dets_pred[i],
calib[i],
opt.center_thresh,
pred,
'bird_pred_gt',
img_id='out')
# debugger.add_img(img, img_id='out')
if opt.debug == 4:
debugger.save_all_imgs(opt.debug_dir,
prefix='{}'.format(iter_id))
else:
debugger.show_all_imgs(pause=True)
def step(split, epoch, opt, data_loader, model, optimizer=None):
if split == 'train':
model.train()
else:
model.eval()
crit = torch.nn.MSELoss()
crit_3d = FusionLoss(opt.device, opt.weight_3d, opt.weight_var)
acc_idxs = data_loader.dataset.acc_idxs
edges = data_loader.dataset.edges
edges_3d = data_loader.dataset.edges_3d
shuffle_ref = data_loader.dataset.shuffle_ref
mean = data_loader.dataset.mean
std = data_loader.dataset.std
convert_eval_format = data_loader.dataset.convert_eval_format
Loss, Loss3D = AverageMeter(), AverageMeter()
Acc, MPJPE = AverageMeter(), AverageMeter()
data_time, batch_time = AverageMeter(), AverageMeter()
preds = []
time_str = ''
nIters = len(data_loader)
bar = Bar('{}'.format(opt.exp_id), max=nIters)
end = time.time()
for i, batch in enumerate(data_loader):
data_time.update(time.time() - end)
for k in batch:
if k != 'meta':
batch[k] = batch[k].cuda(device=opt.device, non_blocking=True)
gt_2d = batch['meta']['pts_crop'].cuda(
device=opt.device, non_blocking=True).float() / opt.output_h
output = model(batch['input'])
loss = crit(output[-1]['hm'], batch['target'])
loss_3d = crit_3d(output[-1]['depth'], batch['reg_mask'],
batch['reg_ind'], batch['reg_target'], gt_2d)
for k in range(opt.num_stacks - 1):
loss += crit(output[k], batch['target'])
loss_3d = crit_3d(output[-1]['depth'], batch['reg_mask'],
batch['reg_ind'], batch['reg_target'], gt_2d)
loss += loss_3d
if split == 'train':
optimizer.zero_grad()
loss.backward()
optimizer.step()
else:
input_ = batch['input'].cpu().numpy().copy()
input_[0] = flip(input_[0]).copy()[np.newaxis, ...]
input_flip_var = torch.from_numpy(input_).cuda(device=opt.device,
non_blocking=True)
output_flip_ = model(input_flip_var)
output_flip = shuffle_lr(
flip(output_flip_[-1]['hm'].detach().cpu().numpy()[0]),
shuffle_ref)
output_flip = output_flip.reshape(1, opt.num_output, opt.output_h,
opt.output_w)
output_depth_flip = shuffle_lr(
flip(output_flip_[-1]['depth'].detach().cpu().numpy()[0]),
shuffle_ref)
output_depth_flip = output_depth_flip.reshape(
1, opt.num_output, opt.output_h, opt.output_w)
output_flip = torch.from_numpy(output_flip).cuda(device=opt.device,
non_blocking=True)
output_depth_flip = torch.from_numpy(output_depth_flip).cuda(
device=opt.device, non_blocking=True)
output[-1]['hm'] = (output[-1]['hm'] + output_flip) / 2
output[-1]['depth'] = (output[-1]['depth'] + output_depth_flip) / 2
# pred, amb_idx = get_preds(output[-1]['hm'].detach().cpu().numpy())
# preds.append(convert_eval_format(pred, conf, meta)[0])
Loss.update(loss.item(), batch['input'].size(0))
Loss3D.update(loss_3d.item(), batch['input'].size(0))
Acc.update(
accuracy(output[-1]['hm'].detach().cpu().numpy(),
batch['target'].detach().cpu().numpy(), acc_idxs))
mpeje_batch, mpjpe_cnt = mpjpe(
output[-1]['hm'].detach().cpu().numpy(),
output[-1]['depth'].detach().cpu().numpy(),
batch['meta']['gt_3d'].detach().numpy(),
convert_func=convert_eval_format)
MPJPE.update(mpeje_batch, mpjpe_cnt)
batch_time.update(time.time() - end)
end = time.time()
if not opt.hide_data_time:
time_str = ' |Data {dt.avg:.3f}s({dt.val:.3f}s)' \
' |Net {bt.avg:.3f}s'.format(dt=data_time, bt=batch_time)
Bar.suffix = '{split}: [{0}][{1}/{2}] |Total {total:} |ETA {eta:} '\
'|Loss {loss.avg:.5f} |Loss3D {loss_3d.avg:.5f}'\
'|Acc {Acc.avg:.4f} |MPJPE {MPJPE.avg:.2f}'\
'{time_str}'.format(epoch, i, nIters, total=bar.elapsed_td,
eta=bar.eta_td, loss=Loss, Acc=Acc,
split=split, time_str=time_str,
MPJPE=MPJPE, loss_3d=Loss3D)
if opt.print_iter > 0:
if i % opt.print_iter == 0:
print('{}| {}'.format(opt.exp_id, Bar.suffix))
else:
bar.next()
if opt.debug >= 2:
gt, amb_idx = get_preds(batch['target'].cpu().numpy())
gt *= 4
pred, amb_idx = get_preds(output[-1]['hm'].detach().cpu().numpy())
pred *= 4
debugger = Debugger(ipynb=opt.print_iter > 0, edges=edges)
img = (batch['input'][0].cpu().numpy().transpose(1, 2, 0) * std +
mean) * 256
img = img.astype(np.uint8).copy()
debugger.add_img(img)
debugger.add_mask(
cv2.resize(batch['target'][0].cpu().numpy().max(axis=0),
(opt.input_w, opt.input_h)), img, 'target')
debugger.add_mask(
cv2.resize(
output[-1]['hm'][0].detach().cpu().numpy().max(axis=0),
(opt.input_w, opt.input_h)), img, 'pred')
debugger.add_point_2d(gt[0], (0, 0, 255))
debugger.add_point_2d(pred[0], (255, 0, 0))
debugger.add_point_3d(batch['meta']['gt_3d'].detach().numpy()[0],
'r',
edges=edges_3d)
pred_3d, ignore_idx = get_preds_3d(
output[-1]['hm'].detach().cpu().numpy(),
output[-1]['depth'].detach().cpu().numpy(), amb_idx)
debugger.add_point_3d(convert_eval_format(pred_3d[0]),
'b',
edges=edges_3d)
debugger.show_all_imgs(pause=False)
debugger.show_3d()
bar.finish()
return {
'loss': Loss.avg,
'acc': Acc.avg,
'mpjpe': MPJPE.avg,
'time': bar.elapsed_td.total_seconds() / 60.
}, preds
import numpy as np
from opts import opts
from datasets.dataset.yolo import YOLO
from utils.debugger import Debugger
if __name__ == '__main__':
opt = opts().parse()
dataset = YOLO(opt.data_dir, opt.flip, opt.vflip, opt.rotate, opt.scale,
opt.shear, opt, 'train')
opt = opts().update_dataset_info_and_set_heads(opt, dataset)
for i in range(len(dataset)):
debugger = Debugger(dataset=opt.names)
data = dataset[i]
img = data['input'].transpose(1, 2, 0)
hm = data['hm']
dets_gt = data['meta']['gt_det']
dets_gt[:, :4] *= opt.down_ratio
img = np.clip(((img * dataset.std + dataset.mean) * 255.), 0,
255).astype(np.uint8)
pred = debugger.gen_colormap(hm)
debugger.add_blend_img(img, pred, 'pred_hm')
debugger.add_img(img, img_id='out_pred')
for k in range(len(dets_gt)):
debugger.add_coco_bbox(dets_gt[k, :4],
dets_gt[k, -1],
dets_gt[k, 4],
img_id='out_pred')
debugger.show_all_imgs(pause=True)
def step(split, epoch, opt, data_loader, model, optimizer=None):
if split == 'train':
model.train()
else:
model.eval()
crit = torch.nn.MSELoss()
acc_idxs = data_loader.dataset.acc_idxs
edges = data_loader.dataset.edges
shuffle_ref = data_loader.dataset.shuffle_ref
mean = data_loader.dataset.mean
std = data_loader.dataset.std
convert_eval_format = data_loader.dataset.convert_eval_format
Loss, Acc = AverageMeter(), AverageMeter()
data_time, batch_time = AverageMeter(), AverageMeter()
preds = []
nIters = len(data_loader)
bar = Bar('{}'.format(opt.exp_id), max=nIters)
end = time.time()
for i, batch in enumerate(data_loader):
data_time.update(time.time() - end)
input, target, meta = batch['input'], batch['target'], batch['meta']
input_var = input.cuda(device=opt.device, non_blocking=True)
target_var = target.cuda(device=opt.device, non_blocking=True)
output = model(input_var)
loss = crit(output[-1]['hm'], target_var)
for k in range(opt.num_stacks - 1):
loss += crit(output[k], target_var)
if split == 'train':
optimizer.zero_grad()
loss.backward()
optimizer.step()
else:
input_ = input.cpu().numpy().copy()
input_[0] = flip(input_[0]).copy()[np.newaxis, ...]
input_flip_var = torch.from_numpy(input_).cuda(device=opt.device,
non_blocking=True)
output_flip = model(input_flip_var)
output_flip = shuffle_lr(
flip(output_flip[-1]['hm'].detach().cpu().numpy()[0]),
shuffle_ref)
output_flip = output_flip.reshape(1, opt.num_output, opt.output_h,
opt.output_w)
# output_ = (output[-1].detach().cpu().numpy() + output_flip) / 2
output_flip = torch.from_numpy(output_flip).cuda(device=opt.device,
non_blocking=True)
output[-1]['hm'] = (output[-1]['hm'] + output_flip) / 2
pred, conf = get_preds(output[-1]['hm'].detach().cpu().numpy(),
True)
preds.append(convert_eval_format(pred, conf, meta)[0])
Loss.update(loss.detach()[0], input.size(0))
Acc.update(
accuracy(output[-1]['hm'].detach().cpu().numpy(),
target_var.detach().cpu().numpy(), acc_idxs))
batch_time.update(time.time() - end)
end = time.time()
if not opt.hide_data_time:
time_str = ' |Data {dt.avg:.3f}s({dt.val:.3f}s)' \
' |Net {bt.avg:.3f}s'.format(dt = data_time,
bt = batch_time)
else:
time_str = ''
Bar.suffix = '{split}: [{0}][{1}/{2}] |Total {total:} |ETA {eta:}' \
'|Loss {loss.avg:.5f} |Acc {Acc.avg:.4f}'\
'{time_str}'.format(epoch, i, nIters, total=bar.elapsed_td,
eta=bar.eta_td, loss=Loss, Acc=Acc,
split = split, time_str = time_str)
if opt.print_iter > 0:
if i % opt.print_iter == 0:
print('{}| {}'.format(opt.exp_id, Bar.suffix))
else:
bar.next()
if opt.debug >= 2:
gt = get_preds(target.cpu().numpy()) * 4
pred = get_preds(output[-1]['hm'].detach().cpu().numpy()) * 4
debugger = Debugger(ipynb=opt.print_iter > 0, edges=edges)
img = (input[0].numpy().transpose(1, 2, 0) * std + mean) * 256
img = img.astype(np.uint8).copy()
debugger.add_img(img)
debugger.add_mask(
cv2.resize(target[0].numpy().max(axis=0),
(opt.input_w, opt.input_h)), img, 'target')
debugger.add_mask(
cv2.resize(
output[-1]['hm'][0].detach().cpu().numpy().max(axis=0),
(opt.input_w, opt.input_h)), img, 'pred')
debugger.add_point_2d(pred[0], (255, 0, 0))
debugger.add_point_2d(gt[0], (0, 0, 255))
debugger.show_all_imgs(pause=True)
bar.finish()
return {
'loss': Loss.avg,
'acc': Acc.avg,
'time': bar.elapsed_td.total_seconds() / 60.
}, preds
def debug(self, batch, output, iter_id):
cfg = self.cfg
reg = output[3] if cfg.LOSS.REG_OFFSET else None
hm_hp = output[4] if cfg.LOSS.HM_HP else None
hp_offset = output[5] if cfg.LOSS.REG_HP_OFFSET else None
dets = multi_pose_decode(output[0],
output[1],
output[2],
reg=reg,
hm_hp=hm_hp,
hp_offset=hp_offset,
K=cfg.TEST.TOPK)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets[:, :, :4] *= cfg.MODEL.INPUT_RES / cfg.MODEL.OUTPUT_RES
dets[:, :, 5:39] *= cfg.MODEL.INPUT_RES / cfg.MODEL.OUTPUT_RES
dets_gt = batch['meta']['gt_det'].numpy().reshape(1, -1, dets.shape[2])
dets_gt[:, :, :4] *= cfg.MODEL.INPUT_RES / cfg.MODEL.OUTPUT_RES
dets_gt[:, :, 5:39] *= cfg.MODEL.INPUT_RES / cfg.MODEL.OUTPUT_RES
for i in range(1):
debugger = Debugger(dataset=cfg.SAMPLE_METHOD,
ipynb=(cfg.DEBUG == 3),
theme=cfg.DEBUG_THEME)
img = batch['input'][i].detach().cpu().numpy().transpose(1, 2, 0)
img = np.clip(((img * np.array(cfg.DATASET.STD).reshape(
1, 1, 3).astype(np.float32) + cfg.DATASET.MEAN) * 255.), 0,
255).astype(np.uint8)
pred = debugger.gen_colormap(output[0][i].detach().cpu().numpy())
gt = debugger.gen_colormap(batch['hm'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'pred_hm')
debugger.add_blend_img(img, gt, 'gt_hm')
debugger.add_img(img, img_id='out_pred')
for k in range(len(dets[i])):
if dets[i, k, 4] > cfg.MODEL.CENTER_THRESH:
debugger.add_coco_bbox(dets[i, k, :4],
dets[i, k, -1],
dets[i, k, 4],
img_id='out_pred')
debugger.add_coco_hp(dets[i, k, 5:39], img_id='out_pred')
debugger.add_img(img, img_id='out_gt')
for k in range(len(dets_gt[i])):
if dets_gt[i, k, 4] > cfg.MODEL.CENTER_THRESH:
debugger.add_coco_bbox(dets_gt[i, k, :4],
dets_gt[i, k, -1],
dets_gt[i, k, 4],
img_id='out_gt')
debugger.add_coco_hp(dets_gt[i, k, 5:39], img_id='out_gt')
if cfg.LOSS.HM_HP:
pred = debugger.gen_colormap_hp(
output[4][i].detach().cpu().numpy())
gt = debugger.gen_colormap_hp(
batch['hm_hp'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'pred_hmhp')
debugger.add_blend_img(img, gt, 'gt_hmhp')
if cfg.DEBUG == 4:
debugger.save_all_imgs(cfg.LOG_DIR,
prefix='{}'.format(iter_id))
else:
debugger.show_all_imgs(pause=True)
def debug(self, batch, output, iter_id):
opt = self.opt
reg = output['reg'] if opt.reg_offset else None
# print(output)
dets = circledet_decode(output['hm'],
output['cl'],
reg=reg,
cat_spec_wh=opt.cat_spec_wh,
K=opt.K)
# print(dets)
if opt.filter_boarder:
output_h = self.opt.default_resolution[
0] // self.opt.down_ratio #hard coded
output_w = self.opt.default_resolution[
1] // self.opt.down_ratio #hard coded
for i in range(dets.shape[1]):
cp = [0, 0]
cp[0] = dets[0, i, 0]
cp[1] = dets[0, i, 1]
cr = dets[0, i, 2]
if cp[0] - cr < 0 or cp[0] + cr > output_w:
dets[0, i, 3] = 0
continue
if cp[1] - cr < 0 or cp[1] + cr > output_h:
dets[0, i, 3] = 0
continue
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets[:, :, :3] *= opt.down_ratio
dets_gt = batch['meta']['gt_det'].numpy().reshape(1, -1, dets.shape[2])
dets_gt[:, :, :3] *= opt.down_ratio
for i in range(1):
debugger = Debugger(dataset=opt.dataset,
ipynb=(opt.debug == 3),
theme=opt.debugger_theme)
img = batch['input'][i].detach().cpu().numpy().transpose(1, 2, 0)
img = np.clip(((img * opt.std + opt.mean) * 255.), 0,
255).astype(np.uint8)
pred = debugger.gen_colormap(
output['hm'][i].detach().cpu().numpy())
gt = debugger.gen_colormap(batch['hm'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'pred_hm')
debugger.add_blend_img(img, gt, 'gt_hm')
debugger.add_img(img, img_id='out_pred')
for k in range(len(dets[i])):
# print('risk = %f' % dets[i, k, 3])
if dets[i, k, 3] > opt.center_thresh:
debugger.add_coco_circle(dets[i, k, :3],
dets[i, k, -1],
dets[i, k, 3],
img_id='out_pred')
debugger.add_img(img, img_id='out_gt')
for k in range(len(dets_gt[i])):
if dets_gt[i, k, 3] > opt.center_thresh:
debugger.add_coco_circle(dets_gt[i, k, :3],
dets_gt[i, k, -1],
dets_gt[i, k, 3],
img_id='out_gt')
if opt.debug == 4:
debugger.save_all_imgs(opt.debug_dir,
prefix='{}'.format(iter_id))
else:
debugger.show_all_imgs(pause=True)
def kp_detection(db, k_ind, data_aug, debug):
data_rng = system_configs.data_rng
batch_size = system_configs.batch_size
categories = db.configs["categories"]
input_size = db.configs["input_size"]
output_size = db.configs["output_sizes"][0]
border = db.configs["border"]
lighting = db.configs["lighting"]
rand_crop = db.configs["rand_crop"]
rand_color = db.configs["rand_color"]
rand_scales = db.configs["rand_scales"]
gaussian_bump = db.configs["gaussian_bump"]
gaussian_iou = db.configs["gaussian_iou"]
gaussian_rad = db.configs["gaussian_radius"]
max_tag_len = 128
# allocating memory
images = np.zeros((batch_size, 3, input_size[0], input_size[1]),
dtype=np.float32)
t_heatmaps = np.zeros(
(batch_size, categories, output_size[0], output_size[1]),
dtype=np.float32)
l_heatmaps = np.zeros(
(batch_size, categories, output_size[0], output_size[1]),
dtype=np.float32)
b_heatmaps = np.zeros(
(batch_size, categories, output_size[0], output_size[1]),
dtype=np.float32)
r_heatmaps = np.zeros(
(batch_size, categories, output_size[0], output_size[1]),
dtype=np.float32)
ct_heatmaps = np.zeros(
(batch_size, categories, output_size[0], output_size[1]),
dtype=np.float32)
t_regrs = np.zeros((batch_size, max_tag_len, 2), dtype=np.float32)
l_regrs = np.zeros((batch_size, max_tag_len, 2), dtype=np.float32)
b_regrs = np.zeros((batch_size, max_tag_len, 2), dtype=np.float32)
r_regrs = np.zeros((batch_size, max_tag_len, 2), dtype=np.float32)
t_tags = np.zeros((batch_size, max_tag_len), dtype=np.int64)
l_tags = np.zeros((batch_size, max_tag_len), dtype=np.int64)
b_tags = np.zeros((batch_size, max_tag_len), dtype=np.int64)
r_tags = np.zeros((batch_size, max_tag_len), dtype=np.int64)
ct_tags = np.zeros((batch_size, max_tag_len), dtype=np.int64)
tag_masks = np.zeros((batch_size, max_tag_len), dtype=np.uint8)
tag_lens = np.zeros((batch_size, ), dtype=np.int32)
db_size = db.db_inds.size
for b_ind in range(batch_size):
if not debug and k_ind == 0:
db.shuffle_inds()
db_ind = db.db_inds[k_ind]
k_ind = (k_ind + 1) % db_size
# reading image
image_file = db.image_file(db_ind)
image = cv2.imread(image_file)
# reading detections
detections, extreme_pts = db.detections(db_ind)
# cropping an image randomly
if rand_crop:
image, detections, extreme_pts = random_crop_pts(image,
detections,
extreme_pts,
rand_scales,
input_size,
border=border)
else:
assert 0
# image, detections = _full_image_crop(image, detections)
image, detections, extreme_pts = _resize_image_pts(
image, detections, extreme_pts, input_size)
detections, extreme_pts = _clip_detections_pts(image, detections,
extreme_pts)
width_ratio = output_size[1] / input_size[1]
height_ratio = output_size[0] / input_size[0]
# flipping an image randomly
if np.random.uniform() > 0.5:
image[:] = image[:, ::-1, :]
width = image.shape[1]
detections[:, [0, 2]] = width - detections[:, [2, 0]] - 1
extreme_pts[:, :, 0] = width - extreme_pts[:, :, 0] - 1
extreme_pts[:, 1, :], extreme_pts[:, 3, :] = \
extreme_pts[:, 3, :].copy(), extreme_pts[:, 1, :].copy()
image = image.astype(np.float32) / 255.
if not debug:
if rand_color:
color_jittering_(data_rng, image)
if lighting:
lighting_(data_rng, image, 0.1, db.eig_val, db.eig_vec)
normalize_(image, db.mean, db.std)
images[b_ind] = image.transpose((2, 0, 1))
for ind, detection in enumerate(detections):
category = int(detection[-1]) - 1
extreme_pt = extreme_pts[ind]
xt, yt = extreme_pt[0, 0], extreme_pt[0, 1]
xl, yl = extreme_pt[1, 0], extreme_pt[1, 1]
xb, yb = extreme_pt[2, 0], extreme_pt[2, 1]
xr, yr = extreme_pt[3, 0], extreme_pt[3, 1]
xct = (xl + xr) / 2
yct = (yt + yb) / 2
fxt = (xt * width_ratio)
fyt = (yt * height_ratio)
fxl = (xl * width_ratio)
fyl = (yl * height_ratio)
fxb = (xb * width_ratio)
fyb = (yb * height_ratio)
fxr = (xr * width_ratio)
fyr = (yr * height_ratio)
fxct = (xct * width_ratio)
fyct = (yct * height_ratio)
xt = int(fxt)
yt = int(fyt)
xl = int(fxl)
yl = int(fyl)
xb = int(fxb)
yb = int(fyb)
xr = int(fxr)
yr = int(fyr)
xct = int(fxct)
yct = int(fyct)
if gaussian_bump:
width = detection[2] - detection[0]
height = detection[3] - detection[1]
width = math.ceil(width * width_ratio)
height = math.ceil(height * height_ratio)
if gaussian_rad == -1:
radius = gaussian_radius((height, width), gaussian_iou)
radius = max(0, int(radius))
else:
radius = gaussian_rad
draw_gaussian(t_heatmaps[b_ind, category], [xt, yt], radius)
draw_gaussian(l_heatmaps[b_ind, category], [xl, yl], radius)
draw_gaussian(b_heatmaps[b_ind, category], [xb, yb], radius)
draw_gaussian(r_heatmaps[b_ind, category], [xr, yr], radius)
draw_gaussian(ct_heatmaps[b_ind, category], [xct, yct], radius)
else:
t_heatmaps[b_ind, category, yt, xt] = 1
l_heatmaps[b_ind, category, yl, xl] = 1
b_heatmaps[b_ind, category, yb, xb] = 1
r_heatmaps[b_ind, category, yr, xr] = 1
tag_ind = tag_lens[b_ind]
t_regrs[b_ind, tag_ind, :] = [fxt - xt, fyt - yt]
l_regrs[b_ind, tag_ind, :] = [fxl - xl, fyl - yl]
b_regrs[b_ind, tag_ind, :] = [fxb - xb, fyb - yb]
r_regrs[b_ind, tag_ind, :] = [fxr - xr, fyr - yr]
t_tags[b_ind, tag_ind] = yt * output_size[1] + xt
l_tags[b_ind, tag_ind] = yl * output_size[1] + xl
b_tags[b_ind, tag_ind] = yb * output_size[1] + xb
r_tags[b_ind, tag_ind] = yr * output_size[1] + xr
ct_tags[b_ind, tag_ind] = yct * output_size[1] + xct
tag_lens[b_ind] += 1
for b_ind in range(batch_size):
tag_len = tag_lens[b_ind]
tag_masks[b_ind, :tag_len] = 1
if debug:
debugger = Debugger(num_classes=1)
t_hm = debugger.gen_colormap(t_heatmaps[0])
l_hm = debugger.gen_colormap(l_heatmaps[0])
b_hm = debugger.gen_colormap(b_heatmaps[0])
r_hm = debugger.gen_colormap(r_heatmaps[0])
ct_hm = debugger.gen_colormap(ct_heatmaps[0])
img = images[0] * db.std.reshape(3, 1, 1) + db.mean.reshape(3, 1, 1)
img = (img * 255).astype(np.uint8).transpose(1, 2, 0)
debugger.add_blend_img(img, t_hm, 't_hm')
debugger.add_blend_img(img, l_hm, 'l_hm')
debugger.add_blend_img(img, b_hm, 'b_hm')
debugger.add_blend_img(img, r_hm, 'r_hm')
debugger.add_blend_img(
img, np.maximum(np.maximum(t_hm, l_hm), np.maximum(b_hm, r_hm)),
'extreme')
debugger.add_blend_img(img, ct_hm, 'center')
debugger.show_all_imgs(pause=True)
images = torch.from_numpy(images)
t_heatmaps = torch.from_numpy(t_heatmaps)
l_heatmaps = torch.from_numpy(l_heatmaps)
b_heatmaps = torch.from_numpy(b_heatmaps)
r_heatmaps = torch.from_numpy(r_heatmaps)
ct_heatmaps = torch.from_numpy(ct_heatmaps)
t_regrs = torch.from_numpy(t_regrs)
l_regrs = torch.from_numpy(l_regrs)
b_regrs = torch.from_numpy(b_regrs)
r_regrs = torch.from_numpy(r_regrs)
t_tags = torch.from_numpy(t_tags)
l_tags = torch.from_numpy(l_tags)
b_tags = torch.from_numpy(b_tags)
r_tags = torch.from_numpy(r_tags)
ct_tags = torch.from_numpy(ct_tags)
tag_masks = torch.from_numpy(tag_masks)
return {
"xs": [images, t_tags, l_tags, b_tags, r_tags, ct_tags],
"ys": [
t_heatmaps, l_heatmaps, b_heatmaps, r_heatmaps, ct_heatmaps,
tag_masks, t_regrs, l_regrs, b_regrs, r_regrs
]
}, k_ind
def debug(self, batch, output, iter_id):
opt = self.opt
# reg = output['reg'] if opt.reg_offset else None
reg = output['reg'][0:1] if opt.reg_offset else None
# dets = ctdet_decode(
# output['hm'], output['wh'], reg=reg,
# cat_spec_wh=opt.cat_spec_wh, K=opt.K)
dets = ctdet_decode(output['hm'][0:1],
output['wh'][0:1],
reg=reg,
cat_spec_wh=opt.cat_spec_wh,
K=opt.K)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets[:, :, :4] *= opt.down_ratio
# FIXME: change from tensor to list and then reshape
# dets_gt = batch['meta']['gt_det'].numpy().reshape(1, -1, dets.shape[2])
# batch['meta_gt_det'] = [128, 128, 6]
gt_det = batch['meta_gt_det'][0:1]
gt_det = np.array(gt_det, dtype=np.float32) if len(gt_det) > 0 else \
np.zeros((1, 6), dtype=np.float32)
dets_gt = gt_det.reshape(1, -1, dets.shape[2])
# print(batch['meta_img_id'][0:1])
dets_gt[:, :, :4] *= opt.down_ratio
for i in range(1):
debugger = Debugger(dataset=opt.dataset,
ipynb=(opt.debug == 3),
theme=opt.debugger_theme)
img = batch['input'][i].detach().cpu().numpy().transpose(1, 2, 0)
img = np.clip(((img * opt.std + opt.mean) * 255.), 0,
255).astype(np.uint8)
pred = debugger.gen_colormap(
output['hm'][i].detach().cpu().numpy())
gt = debugger.gen_colormap(batch['hm'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, 'pred_hm')
debugger.add_blend_img(img, gt, 'gt_hm')
debugger.add_img(img, img_id='out_pred')
for k in range(len(dets[i])):
if dets[i, k, 4] > opt.center_thresh:
debugger.add_coco_bbox(dets[i, k, :4],
dets[i, k, -1],
dets[i, k, 4],
img_id='out_pred')
debugger.add_img(img, img_id='out_gt')
for k in range(len(dets_gt[i])):
if dets_gt[i, k, 4] > opt.center_thresh:
debugger.add_coco_bbox(dets_gt[i, k, :4],
dets_gt[i, k, -1],
dets_gt[i, k, 4],
img_id='out_gt')
if opt.debug == 4:
debugger.save_all_imgs(opt.debug_dir,
prefix='{}'.format(iter_id))
elif opt.debug == 5:
debugger.show_all_imgs(pause=opt.pause,
logger=self.logger,
step=iter_id)
else:
debugger.show_all_imgs(pause=opt.pause, step=iter_id)
def debug(self, batch, output, iter_id):
opt = self.opt
reg = output['reg'] if opt.reg_offset else None
dets = rodet_decode(output['hm'],
output['wh'],
output['angle'],
reg=reg,
cat_spec_wh=opt.cat_spec_wh,
cat_spec_angle=opt.cat_spec_angle,
K=opt.K)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets[:, :, :4] *= opt.down_ratio
# dets_gt_dense = rodet_decode(
# batch['hm'], batch['dense_wh'], batch['dense_angle'], reg=reg,
# cat_spec_wh=opt.cat_spec_wh, cat_spec_angle=opt.cat_spec_angle, K=opt.K)
# dets_gt_dense = dets_gt_dense.detach().cpu().numpy().reshape(1, -1, dets_gt_dense.shape[2])
# dets_gt_dense[:, :, :4] *= opt.down_ratio
dets_gt = batch['meta']['gt_det'].numpy().reshape(1, -1, dets.shape[2])
img_name = batch['meta']['img_name']
dets_gt[:, :, :4] *= opt.down_ratio
for i in range(1):
debugger = Debugger(dataset=opt.dataset,
ipynb=(opt.debug == 3),
theme=opt.debugger_theme)
img = batch['input'][i].detach().cpu().numpy().transpose(1, 2, 0)
img = np.clip(((img * opt.std + opt.mean) * 255.), 0,
255).astype(np.uint8)
pred = debugger.gen_colormap(
output['hm'][i].detach().cpu().numpy())
# gt_ang_mask = debugger.gen_colormap(batch['dense_angle_mask'][i].detach().cpu().numpy())
# gt_ang = debugger.gen_colormap(batch['dense_angle'][i].detach().cpu().numpy())
gt = debugger.gen_colormap(batch['hm'][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, '{}_pred_hm'.format(img_name))
# debugger.add_blend_img(img, gt_ang_mask, 'gt_angle_mask')
# debugger.add_blend_img(img, gt_ang, 'gt_angle')
debugger.add_blend_img(img, gt, '{}_gt_hm'.format(img_name))
debugger.add_img(img, img_id='{}_out_pred'.format(img_name))
for k in range(len(dets[i])):
if dets[i, k, 5] > opt.center_thresh:
# print("pred dets add_rbbox=======================")
debugger.add_rbbox(dets[i, k, :5],
dets[i, k, -1],
dets[i, k, 5],
show_txt=False,
img_id='{}_out_pred'.format(img_name))
# debugger.add_img(img, img_id='{}_dets_gt_dense'.format(img_name))
# for k in range(len(dets_gt_dense[i])):
# if dets_gt_dense[i, k, 5] > opt.center_thresh:
# # print("pred dets add_rbbox=======================")
# debugger.add_rbbox(dets_gt_dense[i, k, :5], dets_gt_dense[i, k, -1],
# dets_gt_dense[i, k, 5], show_txt=False, img_id='{}_dets_gt_dense'.format(img_name))
debugger.add_img(img, img_id='{}_out_gt'.format(img_name))
for k in range(len(dets_gt[i])):
if dets_gt[i, k, 5] > opt.center_thresh:
# print("GT add_rbbox=======================")
# 说明add_rbbox(self, rbbox, cat, conf=1, show_txt=True, img_id='default')
# gt格式 gt_det.append([ct[0], ct[1], w, h, a, 1, cls_id])
debugger.add_rbbox(dets_gt[i, k, :5],
dets_gt[i, k, -1],
dets_gt[i, k, 5],
show_txt=False,
img_id='{}_out_gt'.format(img_name))
if opt.debug == 4:
debugger.save_all_imgs(opt.debug_dir,
prefix='{}'.format(iter_id))
else:
debugger.show_all_imgs(pause=True)
def debug(self, batch, output, iter_id, dataset):
opt = self.opt
if "pre_hm" in batch:
output.update({"pre_hm": batch["pre_hm"]})
dets = generic_decode(output, K=opt.K, opt=opt)
for k in dets:
dets[k] = dets[k].detach().cpu().numpy()
dets_gt = batch["meta"]["gt_det"]
for i in range(1):
debugger = Debugger(opt=opt, dataset=dataset)
img = batch["image"][i].detach().cpu().numpy().transpose(1, 2, 0)
img = np.clip(((img * dataset.std + dataset.mean) * 255.0), 0, 255).astype(np.uint8)
pred = debugger.gen_colormap(output["hm"][i].detach().cpu().numpy())
gt = debugger.gen_colormap(batch["hm"][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, "pred_hm")
debugger.add_blend_img(img, gt, "gt_hm")
if "pre_img" in batch:
pre_img = batch["pre_img"][i].detach().cpu().numpy().transpose(1, 2, 0)
pre_img = np.clip(((pre_img * dataset.std + dataset.mean) * 255), 0, 255).astype(np.uint8)
debugger.add_img(pre_img, "pre_img_pred")
debugger.add_img(pre_img, "pre_img_gt")
if "pre_hm" in batch:
pre_hm = debugger.gen_colormap(batch["pre_hm"][i].detach().cpu().numpy())
debugger.add_blend_img(pre_img, pre_hm, "pre_hm")
debugger.add_img(img, img_id="out_pred")
if "ltrb_amodal" in opt.heads:
debugger.add_img(img, img_id="out_pred_amodal")
debugger.add_img(img, img_id="out_gt_amodal")
# Predictions
for k in range(len(dets["scores"][i])):
if dets["scores"][i, k] > opt.vis_thresh:
debugger.add_coco_bbox(
dets["bboxes"][i, k] * opt.down_ratio,
dets["clses"][i, k],
dets["scores"][i, k],
img_id="out_pred",
)
if "ltrb_amodal" in opt.heads:
debugger.add_coco_bbox(
dets["bboxes_amodal"][i, k] * opt.down_ratio,
dets["clses"][i, k],
dets["scores"][i, k],
img_id="out_pred_amodal",
)
if "hps" in opt.heads and int(dets["clses"][i, k]) == 0:
debugger.add_coco_hp(dets["hps"][i, k] * opt.down_ratio, img_id="out_pred")
if "tracking" in opt.heads:
debugger.add_arrow(
dets["cts"][i][k] * opt.down_ratio,
dets["tracking"][i][k] * opt.down_ratio,
img_id="out_pred",
)
debugger.add_arrow(
dets["cts"][i][k] * opt.down_ratio,
dets["tracking"][i][k] * opt.down_ratio,
img_id="pre_img_pred",
)
# Ground truth
debugger.add_img(img, img_id="out_gt")
for k in range(len(dets_gt["scores"][i])):
if dets_gt["scores"][i][k] > opt.vis_thresh:
debugger.add_coco_bbox(
dets_gt["bboxes"][i][k] * opt.down_ratio,
dets_gt["clses"][i][k],
dets_gt["scores"][i][k],
img_id="out_gt",
)
if "ltrb_amodal" in opt.heads:
debugger.add_coco_bbox(
dets_gt["bboxes_amodal"][i, k] * opt.down_ratio,
dets_gt["clses"][i, k],
dets_gt["scores"][i, k],
img_id="out_gt_amodal",
)
if "hps" in opt.heads and (int(dets["clses"][i, k]) == 0):
debugger.add_coco_hp(dets_gt["hps"][i][k] * opt.down_ratio, img_id="out_gt")
if "tracking" in opt.heads:
debugger.add_arrow(
dets_gt["cts"][i][k] * opt.down_ratio,
dets_gt["tracking"][i][k] * opt.down_ratio,
img_id="out_gt",
)
debugger.add_arrow(
dets_gt["cts"][i][k] * opt.down_ratio,
dets_gt["tracking"][i][k] * opt.down_ratio,
img_id="pre_img_gt",
)
if "hm_hp" in opt.heads:
pred = debugger.gen_colormap_hp(output["hm_hp"][i].detach().cpu().numpy())
gt = debugger.gen_colormap_hp(batch["hm_hp"][i].detach().cpu().numpy())
debugger.add_blend_img(img, pred, "pred_hmhp")
debugger.add_blend_img(img, gt, "gt_hmhp")
if "rot" in opt.heads and "dim" in opt.heads and "dep" in opt.heads:
dets_gt = {k: dets_gt[k].cpu().numpy() for k in dets_gt}
calib = batch["meta"]["calib"].detach().numpy() if "calib" in batch["meta"] else None
det_pred = generic_post_process(
opt,
dets,
batch["meta"]["c"].cpu().numpy(),
batch["meta"]["s"].cpu().numpy(),
output["hm"].shape[2],
output["hm"].shape[3],
self.opt.num_classes,
calib,
)
det_gt = generic_post_process(
opt,
dets_gt,
batch["meta"]["c"].cpu().numpy(),
batch["meta"]["s"].cpu().numpy(),
output["hm"].shape[2],
output["hm"].shape[3],
self.opt.num_classes,
calib,
)
debugger.add_3d_detection(
batch["meta"]["img_path"][i],
batch["meta"]["flipped"][i],
det_pred[i],
calib[i],
vis_thresh=opt.vis_thresh,
img_id="add_pred",
)
debugger.add_3d_detection(
batch["meta"]["img_path"][i],
batch["meta"]["flipped"][i],
det_gt[i],
calib[i],
vis_thresh=opt.vis_thresh,
img_id="add_gt",
)
debugger.add_bird_views(det_pred[i], det_gt[i], vis_thresh=opt.vis_thresh, img_id="bird_pred_gt")
if opt.debug == 4:
debugger.save_all_imgs(opt.debug_dir, prefix="{}".format(iter_id))
else:
debugger.show_all_imgs(pause=True)
