def evaluate_crf(model, x_data, y_data):
total = []
for x, y in tqdm(zip(x_data, y_data), total=len(x_data)):
x = read_image(x)
y = read_mask(y)
y_pred = model.predict(x)[0] # > THRESHOLD
y_pred = y_pred.astype(np.float32)
y_pred = apply_crf(x[0] * 255, y_pred)
total.append(y_pred)
return np.array(total)
def evaluate_crf(model, x_data, y_data):
total = []
for x, y in tqdm(zip(x_data, y_data), total=len(x_data)):
x = read_image(x)
y = read_mask(y)
y_pred = model.predict(x)[0] > 0.5
y_pred = y_pred.astype(np.float32)
y_pred = apply_crf(x[0]*255, y_pred)
value = get_metrics(y, y_pred)
total.append(value)
mean_value = np.mean(total, axis=0)
print(mean_value)
def save_images(model, x_data, y_data):
for i, (x, y) in tqdm(enumerate(zip(x_data, y_data)), total=len(x_data)):
x = read_image(x)
y = read_mask(y)
## Prediction
y_pred_baseline = model.predict(x)[0] > 0.5
y_pred_crf = apply_crf(x[0] * 255, y_pred_baseline.astype(np.float32))
y_pred_tta = tta_model(model, x[0]) > 0.5
y_pred_tta_crf = apply_crf(x[0] * 255, y_pred_tta.astype(np.float32))
y_pred_crf = np.expand_dims(y_pred_crf, axis=-1)
y_pred_tta_crf = np.expand_dims(y_pred_tta_crf, axis=-1)
sep_line = np.ones((256, 10, 3)) * 255
## MeanIoU
miou_baseline = get_mean_iou(y, y_pred_baseline)
miou_crf = get_mean_iou(y, y_pred_crf)
miou_tta = get_mean_iou(y, y_pred_tta)
miou_tta_crf = get_mean_iou(y, y_pred_tta_crf)
print(miou_baseline, miou_crf, miou_crf, miou_tta_crf)
y1 = mask_to_3d(y) * 255
y2 = mask_to_3d(y_pred_baseline) * 255.0
y3 = mask_to_3d(y_pred_crf) * 255.0
y4 = mask_to_3d(y_pred_tta) * 255.0
y5 = mask_to_3d(y_pred_tta_crf) * 255.0
# y2 = cv2.putText(y2, str(miou_baseline), (0, 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (255, 0, 0), 1)
all_images = [
x[0] * 255, sep_line, y1, sep_line, y2, sep_line, y3, sep_line, y4,
sep_line, y5
]
cv2.imwrite(f"results/{i}.png", np.concatenate(all_images, axis=1))
with torch.no_grad():
# Start time
start_time = time.time()
## Prediction
pred_y, pred_m = model(image)
pred_y = torch.sigmoid(pred_y)
pred_m = torch.sigmoid(pred_m)
mask = pred_y
# End timer
end_time = time.time() - start_time
time_taken.append(end_time)
print("{} - {:.10f}".format(image_name, end_time))
mask = mask[0].cpu().numpy()
mask = np.squeeze(mask, axis=0)
mask = mask > 0.5
mask = mask.astype(np.float32)
mask = cv2.resize(mask, (W, H))
mask = apply_crf(ori_image, mask)
mask = mask * 255.0
mask_path = os.path.join(MASK_PATH, image_name)
cv2.imwrite(mask_path, mask)
mean_time_taken = np.mean(time_taken)
mean_fps = 1 / mean_time_taken
print("Mean FPS: ", mean_fps)
import argparse
from scipy import misc
import caffe
import tempfile
from math import ceil
parser = argparse.ArgumentParser()
parser.add_argument('listfile', help='one line should contain paths "img0.ext img1.ext gt.flo out.flo"')
args = parser.parse_args()
if(not os.path.exists(args.listfile)): raise BaseException('listfile does not exist: '+args.listfile)
def readTupleList(filename):
list = []
for line in open(filename).readlines():
if line.strip() != '':
list.append(line.split())
return list
ops = readTupleList(args.listfile)
for ent in ops:
fn_img = ent[0]
fn_flo = ent[3]
fn_out = fn_flo[:fn_flo.rindex('.')] + '_crf.flo'
crf_flo = apply_crf(fn_img, fn_flo)
write_flow(fn_out, crf_flo)
print("Wrote to: " + fn_out)