def _process_batch(sess, original_images, semantic_predictions, image_names,
image_heights, image_widths, image_id_offset, save_dir,
raw_save_dir, train_id_to_eval_id=None):
"""Evaluates one single batch qualitatively.
Args:
sess: TensorFlow session.
original_images: One batch of original images.
semantic_predictions: One batch of semantic segmentation predictions.
image_names: Image names.
image_heights: Image heights.
image_widths: Image widths.
image_id_offset: Image id offset for indexing images.
save_dir: The directory where the predictions will be saved.
raw_save_dir: The directory where the raw predictions will be saved.
train_id_to_eval_id: A list mapping from train id to eval id.
"""
(original_images,
semantic_predictions,
image_names,
image_heights,
image_widths) = sess.run([original_images, semantic_predictions,
image_names, image_heights, image_widths])
num_image = semantic_predictions.shape[0]
for i in range(num_image):
image_height = np.squeeze(image_heights[i])
image_width = np.squeeze(image_widths[i])
original_image = np.squeeze(original_images[i])
semantic_prediction = np.squeeze(semantic_predictions[i])
crop_semantic_prediction = semantic_prediction[:image_height, :image_width]
# Save image.
save_annotation.save_annotation(
original_image, save_dir, _IMAGE_FORMAT % (image_id_offset + i),
add_colormap=False)
# Save prediction.
save_annotation.save_annotation(
crop_semantic_prediction, save_dir,
_PREDICTION_FORMAT % (image_id_offset + i), add_colormap=True,
colormap_type=FLAGS.colormap_type)
if FLAGS.also_save_raw_predictions:
image_filename = os.path.basename(image_names[i])
if train_id_to_eval_id is not None:
crop_semantic_prediction = _convert_train_id_to_eval_id(
crop_semantic_prediction,
train_id_to_eval_id)
save_annotation.save_annotation(
crop_semantic_prediction, raw_save_dir, image_filename,
add_colormap=False)
image_center = (2.0 / 255) * image_rgb - 1.0
image_center = image_center.astype(np.float32)
images = np.expand_dims(image_center, axis=0)
images_pth = np.expand_dims(np.transpose(image_center, axes=(2, 0, 1)),
axis=0)
images_pth = torch.from_numpy(images_pth).to(device)
deeplab = model.deeplab(num_classes=21,
pretrained=True,
atrous_rates=val_atrous_rates,
output_stride=val_output_stride,
checkpoint_path=checkpoint_path).to(device)
deeplab.eval()
with torch.no_grad():
logits = deeplab(images_pth)
logits = torch.nn.functional.softmax(logits, dim=1)
#logits = model.resize_bilinear(logits, original_size)
labels = torch.argmax(logits, dim=1)
label_np = np.squeeze(labels.cpu().numpy()).astype(np.uint8)
label_np = unpad_and_resize(label_np, original_size)
output_path = image_path.replace('.jpg', '_seg')
file_name = output_path.split('/')[-1]
save_dir = output_path.replace(file_name, '')
save_annotation.save_annotation(label_np,
save_dir,
file_name,
add_colormap=True,
colormap_type=colormap_type)
print('The result is saved to: ', save_dir)
def _process_batch(sess,
original_images,
semantic_predictions,
image_names,
image_heights,
image_widths,
image_id_offset,
save_dir,
raw_save_dir,
train_id_to_eval_id=None,
input_feature,
middle_feature,
aspp):
"""Evaluates one single batch qualitatively.
Args:
sess: TensorFlow session.
original_images: One batch of original images.
semantic_predictions: One batch of semantic segmentation predictions.
image_names: Image names.
image_heights: Image heights.
image_widths: Image widths.
image_id_offset: Image id offset for indexing images.
save_dir: The directory where the predictions will be saved.
raw_save_dir: The directory where the raw predictions will be saved.
train_id_to_eval_id: A list mapping from train id to eval id.
"""
(original_images, semantic_predictions, image_names, image_heights,
image_widths, input_feature, middle_feature, aspp[0], aspp[1], aspp[2],
aspp[3]) = sess.run([
original_images, semantic_predictions, image_names, image_heights,
image_widths, input_feature, middle_feature, aspp[0], aspp[1],
aspp[2], aspp[3]
])
num_image = semantic_predictions.shape[0]
for i in range(num_image):
image_height = np.squeeze(image_heights[i])
image_width = np.squeeze(image_widths[i])
original_image = np.squeeze(original_images[i])
semantic_prediction = np.squeeze(semantic_predictions[i])
crop_semantic_prediction = semantic_prediction[:image_height, :
image_width]
input_feat = np.squeeze(input_feature[i])
middle_feat = np.squeeze(middle_feature[i])
aspp0_feat = np.squeeze(aspp[0][i])
aspp1_feat = np.squeeze(aspp[1][i])
aspp2_feat = np.squeeze(aspp[2][i])
aspp3_feat = np.squeeze(aspp[3][i])
num_channel = input_feat[-1]
for j in range(num_channel):
_save_annotation(input_feat[:, :, j], save_dir + '_input',
_FEAT_FORMAT % (image_id_offset + i, j))
num_channel = middle_feat[-1]
for j in range(num_channel):
_save_annotation(middle_feat[:, :, j], save_dir + '_middle',
_FEAT_FORMAT % (image_id_offset + i, j))
num_channel = aspp0_feat[-1]
for j in range(num_channel):
_save_annotation(aspp0_feat[:, :, j], save_dir + '_aspp0',
_FEAT_FORMAT % (image_id_offset + i, j))
num_channel = aspp1_feat[-1]
for j in range(num_channel):
_save_annotation(aspp1_feat[:, :, j], save_dir + '_aspp1',
_FEAT_FORMAT % (image_id_offset + i, j))
num_channel = aspp2_feat[-1]
for j in range(num_channel):
_save_annotation(aspp2_feat[:, :, j], save_dir + '_aspp2',
_FEAT_FORMAT % (image_id_offset + i, j))
num_channel = aspp3_feat[-1]
for j in range(num_channel):
_save_annotation(aspp3_feat[:, :, j], save_dir + '_aspp3',
_FEAT_FORMAT % (image_id_offset + i, j))
# Save image.
save_annotation.save_annotation(original_image,
save_dir,
_IMAGE_FORMAT % (image_id_offset + i),
add_colormap=False)
# Save prediction.
save_annotation.save_annotation(crop_semantic_prediction,
save_dir,
_PREDICTION_FORMAT %
(image_id_offset + i),
add_colormap=True,
colormap_type=FLAGS.colormap_type)
if FLAGS.also_save_raw_predictions:
image_filename = os.path.basename(image_names[i])
if train_id_to_eval_id is not None:
crop_semantic_prediction = _convert_train_id_to_eval_id(
crop_semantic_prediction, train_id_to_eval_id)
save_annotation.save_annotation(crop_semantic_prediction,
raw_save_dir,
image_filename,
add_colormap=False)
def _process_batch(sess,
original_images,
semantic_predictions,
image_names,
image_heights,
image_widths,
image_id_offset,
save_dir,
raw_save_dir,
logits,
train_id_to_eval_id=None):
"""Evaluates one single batch qualitatively.
Args:
sess: TensorFlow session.
original_images: One batch of original images.
semantic_predictions: One batch of semantic segmentation predictions.
image_names: Image names.
image_heights: Image heights.
image_widths: Image widths.
image_id_offset: Image id offset for indexing images.
save_dir: The directory where the predictions will be saved.
raw_save_dir: The directory where the raw predictions will be saved.
train_id_to_eval_id: A list mapping from train id to eval id.
"""
(original_images, semantic_predictions, image_names, image_heights,
image_widths, logits) = sess.run([
original_images, semantic_predictions, image_names, image_heights,
image_widths, logits
])
num_image = semantic_predictions.shape[0]
for i in range(num_image):
image_height = np.squeeze(image_heights[i])
image_width = np.squeeze(image_widths[i])
original_image = np.squeeze(original_images[i])
semantic_prediction = np.squeeze(semantic_predictions[i])
crop_semantic_prediction = semantic_prediction[:image_height, :
image_width]
#np.save(FLAGS.logits_file+"/"+str(image_names[i])+".npy",sess.run(logits))
#print("logits saved")
# Save image.
save_annotation.save_annotation(original_image,
save_dir,
"%s_image" % (image_names[i]),
add_colormap=False)
# Uncomment if you want to save logits along with visualization examples
#path = FLAGS.logits_file+"/"+ str(image_names[i])
#np.save(path,logits)
#print("logits saved")
#Save ground truth in the vis/segmentation_results folder
# gt_dir = "/home/pf/pfshare/data/MA_Rajanie/models/research/deeplab/datasets/lake/SegmentationClassPNG"
# gt = str(image_names[i]).replace("b", "").replace("'", "")+".png"
# if gt in gt_dir:
# shutil.copy(file, save_dir)
# Save prediction.
save_annotation.save_annotation(crop_semantic_prediction,
save_dir,
"%s_pred" % (image_names[i]),
add_colormap=True,
colormap_type=FLAGS.colormap_type)
if FLAGS.also_save_raw_predictions:
image_filename = os.path.basename(image_names[i])
if train_id_to_eval_id is not None:
crop_semantic_prediction = _convert_train_id_to_eval_id(
crop_semantic_prediction, train_id_to_eval_id)
save_annotation.save_annotation(crop_semantic_prediction,
raw_save_dir,
image_filename,
add_colormap=False)
