def get_batch(self):
idx_video = random.randint(0, self.video_list.__len__() - 1)
idx_frame = random.randint(1, 100 - self.T_in)
lr_dir = self.trainset_dir + '/' + self.video_list[idx_video] + '/lr_x' + str(self.upscale_factor) + '_BI'
hr_dir = self.trainset_dir + '/' + self.video_list[idx_video] + '/hr'
# read HR & LR frames
L_frames = []
for i in range(self.T_in):
L_frames.append(LoadImage(lr_dir + '/lr' + str(idx_frame + i) + '.bmp'))
H_frames = LoadImage(hr_dir + '/hr' + str(idx_frame + self.T_in // 2) + '.bmp')
L_frames = np.asarray(L_frames)
# pad L_frame
L_frames_padded = np.lib.pad(L_frames, pad_width=((self.T_in // 2, self.T_in // 2), (0, 0), (0, 0), (0, 0)), mode='constant')
#H_frames = np.asarray(H_frames[np.newaxis,np.newaxis,:,:,:])
return L_frames, H_frames
def get_y(path):
dir_frames = glob.glob(path + "*.png")
dir_frames.sort()
frames = []
for f in dir_frames:
frames.append(LoadImage(f))
frames = np.asarray(frames)
return frames
def get_x(path):
dir_frames=glob.glob(path+"*.png")
dir_frames.sort()
frames=[]
for f in dir_frames:
frames.append(LoadImage(f))
frames = np.asarray(frames) # print(frames.shape) (20, 100, 115, 3)
frames_padded = np.lib.pad(frames, pad_width=((T_in // 2, T_in // 2), (0, 0), (0, 0), (0, 0)), mode='constant') # print(frames_padded.shape) (26, 100, 115, 3)
return frames,frames_padded
def detect(model,
source,
out,
imgsz,
conf_thres,
iou_thres,
names,
colors=[(255, 30, 0), (50, 0, 255)],
device=torch.device('cpu')):
img, img0 = LoadImage(source, img_size=imgsz)
# Run inference
img, im0 = LoadImage(source, img_size=imgsz)
img = torch.from_numpy(img).to(device)
img = img.float()
img /= 255.0 # 0 - 255 to 0.0 - 1.0
if img.ndimension() == 3:
img = img.unsqueeze(0)
# Inference
with torch.no_grad():
pred = model(img)[0]
# Apply NMS
pred = non_max_suppression(pred, conf_thres, iou_thres)
# Process detections
det = pred[0] # detections
if det is not None and len(det):
# Rescale boxes from img_size to im0 size
det[:, :4] = scale_coords(img.shape[2:], det[:, :4], im0.shape).round()
# Print results
for c in det[:, -1].unique():
n = (det[:, -1] == c).sum() # detections per class
# Write results
for *xyxy, conf, cls in det:
label = '%s %.2f' % (names[int(cls)], conf)
# if cls == 0:
plot_fire(xyxy,
im0,
clas=cls,
label=label,
color=colors[int(cls)],
line_thickness=2)
# Save results (image with detections)
cv2.imwrite(out, im0)
return im0
#LoadParams(sess, [params_G], in_file='params_{}L_x{}.h5'.format(args.L, R))
if args.T == 'G':
# Test using GT videos
avg_psnrs = []
dir_inputs = glob.glob('./inputs/G/*')
for v in dir_inputs:
scene_name = v.split('/')[-1]
os.mkdir('./results/{}L/G/{}/'.format(args.L, scene_name))
dir_frames = glob.glob(v + '/*.png')
dir_frames.sort()
frames = []
for f in dir_frames:
frames.append(LoadImage(f))
frames = np.asarray(frames)
frames_padded = np.lib.pad(frames,
pad_width=((T_in // 2, T_in // 2),
(0, 0), (0, 0), (0, 0)),
mode='constant')
frames_padded = np.lib.pad(frames_padded,
pad_width=((0, 0), (8, 8), (8, 8), (0,
0)),
mode='reflect')
out_Hs = []
for i in range(frames.shape[0]):
print('Scene {}: Frame {}/{} processing'.format(
scene_name, i + 1, frames.shape[0]))
in_H = frames_padded[i:i + T_in] # select T_in frames
randoutputpath = options.randoutputpath
computnormalmasks = options.normal_masks
# default parameters
# pair saliency automatically computes smoothgrad squared
smoothgradsq = False # turn off smoothgradsquared
nsamples_sg = 50 # number of noisy samples to compute
xsteps_ig = 50 # interpolation steps for integrated gradients
stdev_spread_sg = 0.15 # std for smoothgrad noisy samples
gradcam_three_dims = True # gradcam should be 3 channels
# assemble a list of these images
listfiles = tf.io.gfile.listdir(inputimgfolderpath)
demo_batch = []
for fl in listfiles:
demo_batch.append(LoadImage(inputimgfolderpath + fl, resize=True))
demo_batch = np.array(demo_batch)
if logging:
print(demo_batch.shape)
layer_randomization_order = inception_block_names()
# compute normal saliency masks.
if computnormalmasks:
# load of inception model
inception_model = Inceptionv3_Wrapper(chkpointpath=chkpntpath,
lblmetadatapath=labeldatapath)
# specify necessary saliency setup functions.
saliency_dict, n_selector = get_saliency_constructors(
inception_model.graph,