def post_process(self, dets, meta, scale=1):
dets = dets.detach().cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
dets = ctdet_post_process(
dets.copy(), [meta['c']], [meta['s']],
meta['out_height'], meta['out_width'])
return dets
def post_process(self, dets, meta):
dets = dets.reshape(1, -1, dets.shape[-1])
dets = ctdet_post_process(dets.copy(), [meta['c']], [meta['s']],
meta['out_height'], meta['out_width'],
self.opt.num_classes)
for j in range(1, self.opt.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
return dets[0]
def post_process(dets, meta):
dets = dets.detach().cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
dets = ctdet_post_process(dets.copy(), [meta['c']], [meta['s']],meta['out_height'], meta['out_width'], num_classes)
for j in range(1, num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
#print('dets',dets[0].keys())
return dets[0]
def post_process(self, dets, meta): # 是将在feature上的预测结果，映射到原始图像中，给出在原始图像中128个检测框的坐标、及相应置信度
dets = dets.detach().cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
dets = ctdet_post_process(
dets.copy(), [meta['c']], [meta['s']],
meta['out_height'], meta['out_width'], self.opt.num_classes)
for j in range(1, self.opt.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
return dets[0] # dets ::= list, dets[0] ::= dict, key值是1，value维度为128*5
def post_process(self, dets, meta, scale=1):
dets = dets.detach().cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
dets = ctdet_post_process(dets.copy(), [meta['c']], [meta['s']], meta['out_height'], meta['out_width'],
self.opt.num_classes)
for j in range(1, self.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
dets[0][j][:, :4] /= scale
return dets[0]
def post_process(self, dets, meta):
"""
change detection coordinate based on original image size
"""
dets = dets.reshape(1, -1, dets.shape[-1])
dets = ctdet_post_process(dets.copy(), [meta['c']], [meta['s']],
meta['out_height'], meta['out_width'],
self.opt.num_classes)
for j in range(1, self.opt.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
return dets[0]
def save_result(self, output, batch, results):
reg = output['reg'] if self.opt.reg_offset else None
dets = ctdet_decode(
output['hm'], output['wh'], reg=reg,
cat_spec_wh=self.opt.cat_spec_wh, K=self.opt.K)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets_out = ctdet_post_process(
dets.copy(), batch['meta']['c'].cpu().numpy(),
batch['meta']['s'].cpu().numpy(),
output['hm'].shape[2], output['hm'].shape[3], output['hm'].shape[1])
results[batch['meta']['img_id'].cpu().numpy()[0]] = dets_out[0]
def post_process(self, dets, meta, scale=1):
dets = dets.detach().cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
# dets 是一个 list, 每一个元素表示一个 batch item 上的 det,
# det 是一个 dict, key 是 class_id, 从 1 开始, value 是 (n, 5) 的 array, 前 4 个元素表示坐标, 最后一位表示 score
dets = ctdet_post_process(dets.copy(), [meta['c']], [meta['s']],
meta['out_height'], meta['out_width'],
self.opt.num_classes)
for j in range(1, self.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
dets[0][j][:, :4] /= scale
return dets[0]
def post_process(self, dets, meta, scale=1):
"""
recover result to ori img size with meta
"""
dets = dets.detach().cpu().numpy() # [B,K,6] box,score,cls
dets = dets.reshape(1, -1, dets.shape[2]) # [1,B*K,6]
dets = ctdet_post_process( # affine transform
dets.copy(), [meta['c']], [meta['s']], meta['out_height'],
meta['out_width'], self.opt.num_classes)
for j in range(1, self.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
dets[0][j][:, :4] /= scale
return dets[0]
def post_process(self, dets, meta, scale=1, fpn_stride=1):
dets = dets.detach().cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
dets = ctdet_post_process(
dets.copy(), [meta['c']], [meta['s']],
meta['out_height'] / fpn_stride, meta['out_width'] / fpn_stride, self.opt.num_classes)
if self.opt.dataset == 'dota':
coor_len = 8
else:
coor_len = 4
for j in range(1, self.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, coor_len + 1)
dets[0][j][:, :coor_len] /= scale
return dets[0]
def post_process(opt, dets, meta):
dets = dets.detach().cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
dets = ctdet_post_process(
dets.copy(),
[meta["c"]],
[meta["s"]],
meta["out_height"],
meta["out_width"],
opt.num_classes,
)
for j in range(1, opt.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
return dets[0]
def save_result(self, output, batch, results):
reg = output['reg'] if self.opt.reg_offset else None
dets = ctdet_decode(
output['hm'], output['wh'], reg=reg,
cat_spec_wh=self.opt.cat_spec_wh, K=self.opt.K)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets_out = ctdet_post_process(
dets.copy(), batch['meta']['c'].cpu().numpy(),
batch['meta']['s'].cpu().numpy(),
output['hm'].shape[2], output['hm'].shape[3], output['hm'].shape[1])
results[batch['meta']['img_id'].cpu().numpy()[0]] = dets_out[0]
if self.opt.dataset == 'bdd' or self.opt.dataset == 'bddstream':
if 'map_img_id' not in results:
results['map_img_id'] = {}
results['map_img_id'][batch['meta']['img_id'].cpu().numpy()[0]] = batch['meta']['file_name']
def post_process(self, dets, meta, scale=1):
torch.cuda.synchronize()
s1 = time.time()
dets = dets.cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
dets = ctdet_post_process(dets, [meta['c']], [meta['s']],
meta['out_height'], meta['out_width'],
self.num_classes)
for j in range(1, self.num_classes + 1):
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
dets[0][j][:, :4] /= scale
torch.cuda.synchronize()
s2 = time.time()
print("post_process:", s2 - s1)
return dets[0]
def post_process(self, dets, meta, scale=1):
dets = dets.detach().cpu().numpy()
dets = dets.reshape(1, -1, dets.shape[2])
dets = ctdet_post_process(dets.copy(), [meta['c']], [meta['s']],
meta['out_height'], meta['out_width'],
self.opt.num_classes)
for j in range(1, self.num_classes + 1):
l = 5
if self.opt.mdn:
l += 10 if self.opt.flip_test else 8
dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, l)
dets[0][j][:, :4] /= scale
if self.opt.mdn:
if self.opt.flip_test:
dets[0][j][:, 9:15] /= scale
else:
dets[0][j][:, 7:13] /= scale
return dets[0]
def save_result(self, output, batch, results):
reg = output["reg"] if self.opt.reg_offset else None
dets = mot_decode(
output["hm"],
output["wh"],
reg=reg,
cat_spec_wh=self.opt.cat_spec_wh,
K=self.opt.K,
)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets_out = ctdet_post_process(
dets.copy(),
batch["meta"]["c"].cpu().numpy(),
batch["meta"]["s"].cpu().numpy(),
output["hm"].shape[2],
output["hm"].shape[3],
output["hm"].shape[1],
)
results[batch["meta"]["img_id"].cpu().numpy()[0]] = dets_out[0]
def save_result(self, output, batch, results):
reg = output['reg'] if self.opt.reg_offset else None
dets = ctdet_decode(output['hm'],
output['wh'],
reg=reg,
cat_spec_wh=self.opt.cat_spec_wh,
K=1)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets_out = ctdet_post_process(dets.copy(),
batch['meta']['c'].cpu().numpy(),
batch['meta']['s'].cpu().numpy(),
output['hm'].shape[2],
output['hm'].shape[3],
output['hm'].shape[1])
# batch * class * top_K -> dets_out[0][1][0]
# results[batch['meta']['img_id'].cpu().numpy()[0]] = dets_out[0][1][0]
#print(dets[0])
results.append({
'predict': dets[0][0][0:4],
'gt_bbox': batch['meta']['gt_det'][0][0][0:4].tolist()
})
def save_result(self, output, batch, results):
opt = self.opt
if opt.task == 'ctdet':
reg = output['reg'] if self.opt.reg_offset else None
dets = ctdet_decode(output['hm'],
output['wh'],
reg=reg,
cat_spec_wh=self.opt.cat_spec_wh,
K=self.opt.K)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets_out = ctdet_post_process(dets.copy(),
batch['meta']['c'].cpu().numpy(),
batch['meta']['s'].cpu().numpy(),
output['hm'].shape[2],
output['hm'].shape[3],
output['hm'].shape[1])
results[batch['meta']['img_id'].cpu().numpy()[0]] = dets_out[0]
elif opt.task == 'multi_pose':
reg = output['reg'] if self.opt.reg_offset else None
hm_hp = output['hm_hp'] if self.opt.hm_hp else None
hp_offset = output['hp_offset'] if self.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=self.opt.K)
dets = dets.detach().cpu().numpy().reshape(1, -1, dets.shape[2])
dets_out = multi_pose_post_process(
dets.copy(), batch['meta']['c'].cpu().numpy(),
batch['meta']['s'].cpu().numpy(), output['hm'].shape[2],
output['hm'].shape[3])
results[batch['meta']['img_id'].cpu().numpy()[0]] = dets_out[0]
else:
assert 0, 'task not defined!'
def post_process(self, dets_act, meta, scale=1):
dets_act = dets_act.detach().cpu().numpy()
dets_act = dets_act.reshape(1, -1, dets_act.shape[2])
dets_act = ctdet_post_process(dets_act.copy(), [meta['c']],
[meta['s']], meta['out_height'],
meta['out_width'],
self.opt.num_obj_classes,
self.opt.num_act_classes)
# print(dets_act)
# for j in range(1, self.num_obj_classes + 1):
# dets[0][j] = np.array(dets[0][j], dtype=np.float32).reshape(-1, 5)
# dets[0][j][:, :4] /= scale
for j in range(1, self.num_act_classes + 1):
dets_act[0][j] = np.array(dets_act[0][j],
dtype=np.float32).reshape(-1, 7)
dets_act[0][j][:, :6] /= scale
# print(dets_act[0])
return dets_act[0]
