def validate(model):
anchor = generate_anchor(8, [8, ], [0.33, 0.5, 1, 2, 3], 17)
prec1 = 0
model = model.eval()
transform=transforms.Compose([
transforms.ToTensor(),transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225]),
])
lines = []
for k in xrange(20):
all_sample = [ args.vot2018[i] for i in sorted(random.sample(xrange(len(args.vot2018)), 30)) ]
nSamples = len(all_sample)
for i in xrange(nSamples):
sequence = all_sample[i]
ran_id = random.randint(0,len(sequence)-1)
while len(sequence[ran_id])<2:
sequence = all_sample[random.randint(0,nSamples-1)]
ran_id = random.randint(0,len(sequence)-1)
track_obj = sequence[ran_id]
ran_f1 = random.randint(0,len(track_obj)-1)
ran_f2 = random.randint(0,len(track_obj)-1)
lines.append([track_obj[ran_f1],track_obj[ran_f2]])
random.shuffle(lines)
for line in lines:
z,x,gt_box,regression_target,conf_target= load_data(line,0)
inpz = transform(z)
inpx = transform(x)
score, delta = model(inpz.unsqueeze(0).cuda(),inpx.unsqueeze(0).cuda())
delta = delta.permute(1, 2, 3, 0).contiguous().view(4, -1).data.cpu().numpy()
score = F.softmax(score.permute(1, 2, 3, 0).contiguous().view(2, -1), dim=0).data[1, :].cpu().numpy()
delta[0, :] = delta[0, :] * anchor[:, 2] + anchor[:, 0]
delta[1, :] = delta[1, :] * anchor[:, 3] + anchor[:, 1]
delta[2, :] = np.exp(delta[2, :]) * anchor[:, 2]
delta[3, :] = np.exp(delta[3, :]) * anchor[:, 3]
best_pscore_id = np.argmax(score)
target = delta[:, best_pscore_id]
prec1 += bbox_iou(target, gt_box, False)
prec1 = prec1/600
return prec1
def __getitem__(self, index):
assert index <= len(self), 'index range error'
img_path = self.lines[index]
img, target = load_data(img_path, self.train)
if self.transform is not None:
img = self.transform(img)
return img, target
def __getitem__(self, index):
assert index <= len(self), 'index range error'
img_path = self.lines[index]
img, target, smap = load_data(img_path)
img = self.transform(img)
return img, target, smap
def pre_data(train_list, args, train):
print("Pre_load dataset ......")
data_keys = {}
count = 0
for j in range(len(train_list)):
Img_path = train_list[j]
fname = os.path.basename(Img_path)
img, gt_count = load_data(Img_path, args, train)
blob = {}
blob['img'] = img
blob['gt_count'] = gt_count
blob['fname'] = fname
data_keys[count] = blob
count += 1
'''for debug'''
# if j> 10:
# break
return data_keys