np.asarray([db.load_imgs(img) for img in imgs]),
np.asarray([db.get_groundtruth(gt, 255.0) for gt in gts]))
output_file = "output/out{}_{}_{}.png"
head = ['VIDNAME', 'AGE', 'pEPs', 'pCEPs', 'MSSSIM', 'PSNR', 'CQM']
report = CSVLogging("report{}.csv".format(options.method), head)
init_weight = model.get_model().get_weights()
try:
for id in range(db.max_video):
print("VIDEO : {}, {}".format(id, db.videos[id]["input"][-15:]))
model.get_model().set_weights(init_weight)
model.get_model().fit_generator(
generator=get_generator_batched_for_id(id, options.ratio),
samples_per_epoch=min(
int(db.get_count_on_video(id) * options.ratio),
options.max_length),
nb_epoch=max_epoch,
callbacks=[])
if db.get_count_on_video(id) * (
1.0 - options.ratio) > 0 and options.max_length > 0:
max_test = db.get_count_on_video(id) - options.max_length
outputs = model.get_model().predict_generator(
get_generator_test_batched_for_id(id, options.ratio), 5)
gt = db.get_groundtruth_from_id(id)
gt = gt.reshape(list(gt.shape) + [1])
acc = []
for i, output in enumerate(outputs):
cv2.imwrite(
output_file.format(options.method, id, i),
output.reshape([model.output_size, model.output_size]) *
ratio = 0.5
@GeneratorLoop
def get_generator_test_batched_for_id(id):
max_test = db.get_count_on_video(id) - options.max_length
for batch in chunks(db.get_datas_on_one(id), options.batch_size):
imgs, gts = zip(*batch)
yield model.preprocess(
np.asarray([db.load_imgs(img) for img in imgs]),
np.asarray([db.get_groundtruth(gt, 255.0) for gt in gts]))
if options.weight_file is None:
model.get_model().fit_generator(
generator=get_generator_batched_for_id(options.videoid),
samples_per_epoch=int(db.get_count_on_video(options.videoid) * ratio),
nb_epoch=max_epoch)
outputs = model.get_model().predict_generator(
get_generator_batched_for_id(options.videoid),
int(db.get_count_on_video(options.videoid) * (1 - ratio)))
gt = db.get_groundtruth_from_id(options.videoid)
gt = gt.reshape(list(gt.shape) + [1])
acc = []
for i, output in enumerate(outputs):
cv2.imwrite(output_file.format(options.method, i),
output.reshape([model.output_size, model.output_size]) * 255.)
acc += Evaluate(gt, output)
print(acc)
