def check_on_train_clip(video_id, weights, suffix, is_test=False):
if 'resnet' in weights:
model = build_resnet(input_shape)
else:
model = build_model(input_shape)
pool = ThreadPool(processes=8)
model.compile(loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=2.0, pos_cost_multiplier=1.0).compute_loss,
optimizer=Adam(lr=1e-3))
model.load_weights(weights)
model.summary()
priors = priors_from_model(model)
bbox_util = BBoxUtility(NUM_CLASSES, priors)
dataset = SSDDataset(bbox_util=bbox_util, is_test=is_test)
outdir = '../output/predictions_ssd/' + video_id + suffix
os.makedirs(outdir, exist_ok=True)
batch_size = 4
frame_id = 0
for x_batch, frames in dataset.generate_x_for_train_video_id(video_id=video_id, batch_size=batch_size, pool=pool):
predictions = model.predict(x_batch)
results = bbox_util.detection_out(predictions)
for batch_id in range(predictions.shape[0]):
print(results[batch_id])
display_img_with_rects(img=utils.preprocessed_input_to_img_resnet(x_batch[batch_id]) * 255,
results=results,
res_idx=batch_id)
plt.savefig('{}/{:04}.jpg'.format(outdir, frame_id+1))
plt.clf()
frame_id += 1
print(frame_id)
def check(fold, weights):
dataset = ClassificationDataset(fold=fold)
model = build_model_densenet_161()
model.load_weights(weights)
batch_size = 2
for x_batch, y_batch in dataset.generate(batch_size=batch_size):
print(y_batch)
predicted = model.predict_on_batch(x_batch)
print(predicted)
for i in range(batch_size):
plt.imshow(utils.preprocessed_input_to_img_resnet(x_batch[i]))
true_species = y_batch['cat_species'][i]
true_cover = y_batch['cat_cover'][i]
predicted_species = predicted[0][i]
predicted_cover = predicted[1][i]
for cls_id, cls in enumerate(SPECIES_CLASSES):
print('{:12} {:.02f} {:.02f}'.format(
cls, true_species[cls_id], predicted_species[cls_id]))
for cls_id, cls in enumerate(COVER_CLASSES):
print('{:12} {:.02f} {:.02f}'.format(cls, true_cover[cls_id],
predicted_cover[cls_id]))
print(SPECIES_CLASSES[np.argmax(y_batch['cat_species'][i])],
COVER_CLASSES[np.argmax(y_batch['cat_cover'][i])])
plt.show()
def check(weights):
model = build_model(input_shape)
model.compile(loss=MultiboxLoss(NUM_CLASSES, neg_pos_ratio=2.0, pos_cost_multiplier=1.0).compute_loss,
optimizer=Adam(lr=1e-3))
model.load_weights(weights)
model.summary()
priors = priors_from_model(model)
bbox_util = BBoxUtility(NUM_CLASSES, priors)
dataset = SSDDataset(bbox_util=bbox_util)
for x_batch, y_batch in dataset.generate_ssd(batch_size=4, is_training=False, verbose=True, always_shuffle=True):
predictions = model.predict(x_batch)
results = bbox_util.detection_out(predictions)
for batch_id in range(4):
display_img_with_rects(img=utils.preprocessed_input_to_img_resnet(x_batch[batch_id])*255,
results=results,
res_idx=batch_id)
plt.show()
def check_model(model_name, weights, fold):
model = MODELS[model_name].factory(lock_base_model=True)
model.load_weights(weights, by_name=True)
dataset = SingleFrameCNNDataset(
preprocess_input_func=MODELS[model_name].preprocess_input,
batch_size=1,
validation_batch_size=1,
fold=fold)
batch_id = 0
for X, y in dataset.generate_test():
pred = model.predict_on_batch(X)
print()
for i, cls in enumerate(CLASSES):
print(f'gt: {y[0, i]} pred: {pred[0, i]:.03f} {cls}')
batch_id += 1
for batch_frame in range(dataset.batch_size):
plt.imshow(utils.preprocessed_input_to_img_resnet(X[batch_frame]))
# plt.imshow(X[batch_frame]/2.0+0.5)
plt.show()
def check_generator(use_test):
dataset = SingleFrameCNNDataset(
preprocess_input_func=preprocess_input_resnet50,
batch_size=2,
validation_batch_size=2,
fold=1)
batch_id = 0
startTime = time.time()
if use_test:
gen = dataset.generate_test()
else:
gen = dataset.generate()
for X, y in gen:
batch_id += 1
elapsedTime = time.time() - startTime
startTime = time.time()
print(f'{batch_id} {elapsedTime:.3}')
for batch_frame in range(dataset.batch_size):
print(y[batch_frame])
plt.imshow(utils.preprocessed_input_to_img_resnet(X[batch_frame]))
plt.show()
def unprocess_input(x):
return utils.preprocessed_input_to_img_resnet(x)