def test_model(model_name, model, test_loader = validation_set_loader):
model.load_state_dict(torch.load(str(RESULTS_PATH) + "/" + str(model_name) + "/" + str(model_name) + ".pt"))
if USE_CUDA and cuda_available:
model = model.cuda()
model.eval()
preds = []
gts = []
#debug
i = 0
for batch in test_loader:
x = Variable(batch['image'])
if USE_CUDA and cuda_available:
x = x.cuda()
pred = model(x).data.cuda().cpu().numpy().copy()
else:
pred = model(x).data.numpy().copy()
gt = batch['label'].numpy().copy()
preds.append(pred)
gts.append(gt)
# debug
if DEBUG:
if i == 2:
break
else:
i+=1
return np.concatenate(preds), np.concatenate(gts)
def generate_test_da(model_name, model, test_loader=test_set_loader):
model.load_state_dict(torch.load(str(RESULTS_PATH) + "/" + str(model_name) + "/" + str(model_name) + ".pt"))
if USE_CUDA and cuda_available:
model = model.cuda()
model.eval()
preds = []
gts = []
for batch in test_loader:
x = Variable(batch['image'])
if USE_CUDA and cuda_available:
x = x.cuda()
pred = model(x).data.cuda().cpu().numpy().copy()
else:
pred = model(x).data.numpy().copy()
gt = batch['label'].numpy().copy()
preds.append(pred)
gts.append(gt)
output_test = ""
for i in range(len(preds)):
for j in range(len(preds[i])):
for p in range(len(preds[i][j])):
output_test += ","+str(preds[i][j][p])
output_test+="\n"
f = open(model_name+"_output.csv", "w+")
f.write(output_test)
f.close()
def generate_test(model_name, model):
model.load_state_dict(torch.load(str(RESULTS_PATH) + "/" + str(model_name) + "/" + str(model_name) + ".pt"))
test_list = open(TEST_PATH, "r").read().split("\n")
test_list.remove('')
output_test = ""
for img in test_list:
image_path = "images/"+str(img)
im = Image.open(image_path).convert("RGB")
im = transform(im)
if USE_CUDA and cuda_available:
model = model.cuda()
model.eval()
x = Variable(im.unsqueeze(0))
if USE_CUDA and cuda_available:
x = x.cuda()
pred = model(x).data.cuda().cpu().numpy().copy()
else:
pred = model(x).data.numpy().copy()
if not REGRESSION:
idx_max_pred = np.argmax(pred)
idx_classes = idx_max_pred % 16
output_test += str(img)+","+str(idx_classes)+"\n"
else:
output_test += str(img)
for i in pred[0]:
output_test += ","+str(i)
output_test+="\n"
f = open(model_name+"_output.csv", "w+")
f.write(output_test)
f.close()
model = resnet152_model
if args.inp:
print("input: ", args.inp)
image_path = args.inp
im = Image.open(image_path).convert("RGB")
im = transform_test(im)
print(
str(RESULTS_PATH) + "/" + str(model_name) + "/" + str(model_name) +
".pt")
model.load_state_dict(
torch.load(
str(RESULTS_PATH) + "/" + str(model_name) + "/" + str(model_name) +
".pt"))
if USE_CUDA and cuda_available:
model = model.cuda()
model.eval()
x = Variable(im.unsqueeze(0))
if USE_CUDA and cuda_available:
x = x.cuda()
pred = model(x).data.cuda().cpu().numpy().copy()
else:
pred = model(x).data.numpy().copy()
print(pred)
resnet152_model = resnet.resnet152(pretrained=False, **classes)
train_model_iter("resnet152", resnet152_model)
model_name="resnet152"
model=resnet152_model
if args.inp:
print ("input: ", args.inp)
image_path = args.inp
im = Image.open(image_path).convert("RGB")
im = transform(im)
print (str(RESULTS_PATH) + "/" + str(model_name) + "/" + str(model_name) + ".pt")
model.load_state_dict(torch.load(str(RESULTS_PATH) + "/" + str(model_name) + "/" + str(model_name) + ".pt"))
if USE_CUDA and cuda_available:
model = model.cuda()
model.eval()
x = Variable(im.unsqueeze(0))
if USE_CUDA and cuda_available:
x = x.cuda()
pred = model(x).data.cuda().cpu().numpy().copy()
else:
pred = model(x).data.numpy().copy()
print (pred)