def process(self, data_el):
img_desc, ann = data_el
free_name = self.net.get_free_name(img_desc)
new_dataset_name = img_desc.get_res_ds_name()
if self.settings.get('visualize'):
out_meta = self.net.get_result_project_meta()
cls_mapping = {}
for cls_descr in out_meta.classes:
color_s = cls_descr.get('color')
if color_s is not None:
color = color_utils.hex2rgb(color_s)
else:
color = color_utils.get_random_color()
cls_mapping[cls_descr['title']] = color
# hack to draw 'black' regions
cls_mapping = {
k: (1, 1, 1) if max(v) == 0 else v
for k, v in cls_mapping.items()
}
vis_img = self.draw_colored_mask(ann, cls_mapping)
orig_img = img_desc.read_image()
comb_img = imaging.overlay_images(orig_img, vis_img, 0.5)
sep = np.array([[[0, 255, 0]]] * orig_img.shape[0], dtype=np.uint8)
img = np.hstack((orig_img, sep, comb_img))
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
output_img_path = osp.join(self.output_folder, new_dataset_name,
'visualize', free_name + '.png')
os_utils.ensure_base_path(output_img_path)
cv2.imwrite(output_img_path, img)
ann_to_save = deepcopy(ann)
ann_to_save.normalize_figures()
packed_ann = ann_to_save.pack()
dataset_name = img_desc.get_res_ds_name()
if not self.out_project.datasets.has_key(dataset_name):
self.out_project.create_dataset(dataset_name)
out_dataset = self.out_project.datasets.get(dataset_name)
out_item_name = free_name + img_desc.get_image_ext()
# net _always_ downloads images
if img_desc.need_write():
out_dataset.add_item_np(out_item_name,
img_desc.image_data,
ann=packed_ann)
else:
out_dataset.add_item_file(out_item_name,
img_desc.get_img_path(),
ann=packed_ann)
yield ([img_desc, ann], )
def process(self, data_el):
img_desc, ann = data_el
free_name = self.net.get_free_name(img_desc)
new_dataset_name = img_desc.get_res_ds_name()
for out_dir, flag_name, mapping_name in self.odir_flag_mapping:
if not self.settings[flag_name]:
continue
cls_mapping = self.settings[mapping_name]
# hack to draw 'black' regions
if flag_name == 'masks_human':
cls_mapping = {
k: (1, 1, 1) if max(v) == 0 else v
for k, v in cls_mapping.items()
}
img = self.draw_colored_mask(ann, cls_mapping)
if flag_name == 'masks_human':
orig_img = img_desc.read_image()
comb_img = imaging.overlay_images(orig_img, img, 0.5)
sep = np.array([[[0, 255, 0]]] * orig_img.shape[0],
dtype=np.uint8)
img = np.hstack((orig_img, sep, comb_img))
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
output_img_path = osp.join(self.output_folder, new_dataset_name,
out_dir, free_name + '.png')
os_utils.ensure_base_path(output_img_path)
cv2.imwrite(output_img_path, img)
# net _always_ downloads images
if self.settings['images'] is True:
if img_desc.need_write() is True:
self.pr_writer.write_image(img_desc, free_name)
else:
self.pr_writer.copy_image(img_desc, free_name)
if self.settings['annotations'] is True:
ann_to_save = deepcopy(ann)
ann_to_save.normalize_figures()
packed_ann = ann_to_save.pack()
self.pr_writer.write_ann(img_desc, packed_ann, free_name)
yield ([img_desc, ann], )
def run_inference(self):
out_project_fs = copy(self.in_project_fs)
out_project_fs.root_path = self.helper.paths.results_dir
out_project_fs.make_dirs()
inference_feeder = InferenceFeederFactory.create(
self.config, self.helper.in_project_meta,
self.single_image_inference.train_classes)
out_pr_meta = inference_feeder.out_meta
out_pr_meta.to_dir(out_project_fs.project_path)
ia_cnt = out_project_fs.pr_structure.image_cnt
progress = progress_counter.progress_counter_inference(cnt_imgs=ia_cnt)
for sample in self.in_project_fs:
logger.trace('Will process image',
extra={
'dataset_name': sample.ds_name,
'image_name': sample.image_name
})
ann_packed = json_utils.json_load(sample.ann_path)
ann = sly_annotation.Annotation.from_packed(
ann_packed, self.helper.in_project_meta)
img = cv2.imread(sample.img_path)[:, :, ::-1]
res_ann = inference_feeder.feed(
img, ann, self.single_image_inference.inference)
out_ann_fpath = out_project_fs.ann_path(sample.ds_name,
sample.image_name)
res_ann_packed = res_ann.pack()
json_utils.json_dump(res_ann_packed, out_ann_fpath)
if self.debug_copy_images:
out_img_fpath = out_project_fs.img_path(
sample.ds_name, sample.image_name)
os_utils.ensure_base_path(out_img_fpath)
shutil.copy(sample.img_path, out_img_fpath)
progress.iter_done_report()
progress_counter.report_inference_finished()