def make_chip(self, img_name):
image = cv2.imread(osp.join(self.img_dir, img_name))
height, width = image.shape[:2]
img_id = osp.splitext(osp.basename(img_name))[0]
# mask_path = ""
mask_path = osp.join(self.mask_dir, '{}.hdf5'.format(img_id))
with h5py.File(mask_path, 'r') as hf:
mask = np.array(hf['label'])
mask_h, mask_w = mask.shape[:2]
# make chip
region_box, contours = utils.generate_box_from_mask(mask)
region_box = utils.region_postprocess(region_box, contours,
(mask_w, mask_h))
region_box = utils.resize_box(region_box, (mask_w, mask_h),
(width, height))
region_box = utils.generate_crop_region(region_box, (width, height))
try:
region_box = np.vstack(
(region_box, np.array([0, 0, width - 1, height - 1])))
except:
print("empty box")
if args.show:
utils.show_image(image, region_box)
chip_loc = self.write_chip_and_anno(image, img_id, region_box)
return len(region_box), chip_loc
def make_chip(self, sample, imgset):
image = cv2.imread(sample['image'])
height, width = sample['height'], sample['width']
img_id = osp.splitext(osp.basename(sample['image']))[0]
mask_path = osp.join(self.segmentation_dir, '{}.hdf5'.format(img_id))
with h5py.File(mask_path, 'r') as hf:
mask = np.array(hf['label'])
mask_h, mask_w = mask.shape[:2]
# make chip
region_box, contours = utils.generate_box_from_mask(mask)
region_box = utils.region_postprocess(region_box, contours,
(mask_w, mask_h))
region_box = utils.resize_box(region_box, (mask_w, mask_h),
(width, height))
region_box = utils.generate_crop_region(region_box, (width, height))
if args.show:
utils.show_image(image, region_box)
# if imgset == 'train':
# region_box = np.vstack((region_box, np.array([0, 0, width-1, height-1])))
gt_bboxes, gt_cls = sample['bboxes'], sample['cls']
chip_gt_list, chip_label_list, neglect_list = self.generate_region_gt(
region_box, gt_bboxes, gt_cls)
chip_loc = self.write_chip_and_anno(image, img_id, imgset, region_box,
chip_gt_list, chip_label_list,
neglect_list)
return chip_loc
def _boxvis(img, gt_box_list):
img1 = img.copy()
mask = np.zeros_like(img1, dtype=np.uint8)
chip_mask = np.zeros((30, 30, 1), dtype=np.uint8)
vis_mask = np.zeros_like(img1, dtype=np.uint8)
for box in gt_box_list:
cv2.rectangle(img1, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 4)
mask[box[1]:box[3], box[0]:box[2], :] = 255
xmin, ymin, xmax, ymax = np.floor(box / 2048 * 30).astype(np.int)
chip_mask[ymin:ymax + 1, xmin:xmax + 1] = 1
mask_box = utils.generate_box_from_mask(chip_mask)
for box in mask_box:
xmin, ymin, xmax, ymax = np.floor(box / 30 * 2048).astype(np.int)
vis_mask[ymin:ymax + 1, xmin:xmax + 1] = 255
cv2.imwrite('vis1.jpg', img1)
cv2.imwrite('vis2.png', vis_mask)
plt.subplot(1, 2, 1)
plt.imshow(img1[:, :, [2, 1, 0]])
plt.subplot(1, 2, 2)
plt.imshow(vis_mask)
plt.show()
cv2.waitKey()
def make_chip(self, img_name):
image = cv2.imread(osp.join(self.img_dir, img_name))
height, width = image.shape[:2]
img_id = osp.splitext(osp.basename(img_name))[0]
# mask_path = ""
mask_path = osp.join(self.mask_dir, '{}.hdf5'.format(img_id))
with h5py.File(mask_path, 'r') as hf:
mask = np.array(hf['label'])
mask_h, mask_w = mask.shape[:2]
# make chip
region_box, contours = utils.generate_box_from_mask(mask)
# # utils.show_image(mask, np.array(region_box))
region_box = utils.generate_crop_region(region_box, mask, (mask_w, mask_h), (width, height), self.gbm, args.aim)
# # utils.show_image(mask, np.array(region_box))
region_box = utils.resize_box(region_box, (mask_w, mask_h), (width, height))
# if len(region_box) == 0:
# region_box = np.array([[0, 0, width, height]])
# else:
# region_box = np.vstack((region_box, [0, 0, width, height]))
if args.show:
utils.show_image(image[..., ::-1], np.array(region_box))
chip_loc = self.write_chip_and_anno(image, img_id, region_box)
return chip_loc
def main():
if not os.path.exists(dest_datadir):
os.mkdir(dest_datadir)
os.mkdir(image_dir)
os.makedirs(list_dir)
os.mkdir(anno_dir)
train_list = glob.glob(src_traindir + '/*.jpg')
random.shuffle(train_list)
train_list, val_list = train_list[:-2000], train_list[-2000:]
test_list = glob.glob(src_testdir + '/*.jpg')
test_list = [os.path.basename(x)[:-4] for x in test_list]
chip_loc = {}
chip_name_list = []
for imgid in tqdm(test_list):
origin_img = cv2.imread(os.path.join(src_testdir, '%s.jpg' % imgid))
mask_img = cv2.imread(os.path.join(mask_path, '%s.png' % imgid),
cv2.IMREAD_GRAYSCALE)
height, width = mask_img.shape[:2]
mask_box = utils.generate_box_from_mask(mask_img)
mask_box = list(
map(utils.resize_box, mask_box, [(width, height)] * len(mask_box),
[(3200, 1800)] * len(mask_box)))
chip_list = mask_chip(mask_box, (3200, 1800))
# utils._boxvis(cv2.resize(mask_img, (2048, 2048)), chip_list, origin_img)
# cv2.waitKey(0)
for i, chip in enumerate(chip_list):
chip_img = origin_img[chip[1]:chip[3], chip[0]:chip[2], :].copy()
# chip_img = cv2.resize(chip_img, (416, 416), cv2.INTER_AREA)
chip_name = '%s_%d' % (imgid, i)
cv2.imwrite(os.path.join(image_dir, '%s.jpg' % chip_name),
chip_img)
chip_name_list.append(chip_name)
chip_info = {'loc': chip}
chip_loc[chip_name] = chip_info
# write test txt
with open(os.path.join(list_dir, 'test.txt'), 'w') as f:
f.writelines([x + '\n' for x in chip_name_list])
print('write txt.')
# write chip loc json
with open(os.path.join(anno_dir, 'test_chip.json'), 'w') as f:
json.dump(chip_loc, f)
print('write json.')
def main():
label_df = pd.read_csv(annos_path)
label_object = []
detect_object = []
mask_object = []
undetected_img = []
pixel_num = []
for raw_file in tqdm(glob(mask_path + '/*.png')):
img_name = os.path.basename(raw_file)
imgid = os.path.splitext(img_name)[0]
label_file = os.path.join(label_path, img_name)
image_file = os.path.join(image_path, imgid + '.jpg')
mask_img = cv.imread(raw_file, cv.IMREAD_GRAYSCALE)
height, width = mask_img.shape[:2]
pixel_num.append(np.sum(mask_img))
label_box = utils.get_label_box(label_df, imgid + '.jpg')
mask_box = utils.generate_box_from_mask(mask_img)
mask_box = list(
map(utils.resize_box, mask_box, [(width, height)] * len(mask_box),
[(3200, 1800)] * len(mask_box)))
# mask_box = utils.enlarge_box(mask_box, (3200, 1800), ratio=1)
# _boxvis(mask_img, mask_box)
count = 0
for box1 in label_box:
for box2 in mask_box:
if utils.overlap(box2, box1):
count += 1
break
label_object.append(len(label_box))
detect_object.append(count)
mask_object.append(len(mask_box))
if len(label_box) != count:
undetected_img.append(imgid)
print('recall: %f' % (np.sum(detect_object) / np.sum(label_object)))
print('cost avg: %f, std: %f' % (np.mean(pixel_num), np.std(pixel_num)))
print('detect box avg: %f, std %d' %
(np.mean(mask_object), np.std(mask_object)))
def make_chip(self, sample, imgset):
# get image and mask informations
image = cv2.imread(sample['image'])
height, width = sample['height'], sample['width']
img_id = osp.splitext(osp.basename(sample['image']))[0]
mask_path = osp.join(self.segmentation_dir, '{}.hdf5'.format(img_id))
with h5py.File(mask_path, 'r') as hf:
mask = np.array(hf['label'])
mask_h, mask_w = mask.shape[:2]
# make chip
region_box, contours = utils.generate_box_from_mask(mask)
region_box = utils.generate_crop_region(region_box, mask,
(mask_w, mask_h),
(width, height), self.gbm,
args.aim)
region_box = utils.resize_box(region_box, (mask_w, mask_h),
(width, height))
# make tiling
if args.tiling and imgset != "val":
tiling = utils.add_tiling((width, height))
for pattern in tiling:
if utils.iou_calc1(pattern, region_box).max() < 0.85:
region_box = np.vstack((region_box, tiling))
if args.show:
utils.show_image(image[..., ::-1], np.array(region_box))
# get box and class
gt_bboxes, gt_cls = sample['bboxes'], sample['cls']
# generate chip annotations and writer chip image
chip_gt_list, chip_label_list, neglect_list = self.generate_region_gt(
region_box, gt_bboxes, gt_cls)
chip_loc = self.write_chip_and_anno(image, img_id, region_box,
chip_gt_list, chip_label_list,
neglect_list, imgset)
return chip_loc
def _worker(img_path, dataset):
imgid = os.path.basename(img_path)[:-4]
image = cv2.imread(img_path)
height, width = image.shape[:2]
mask_path = os.path.join(segmentation_dir, imgid + '_region.png')
mask_img = cv2.imread(mask_path, cv2.IMREAD_GRAYSCALE)
mask_h, mask_w = mask_img.shape[:2]
region_box, contours = utils.generate_box_from_mask(mask_img)
region_box = utils.region_postprocess(region_box, contours,
(mask_w, mask_h))
region_box = utils.resize_box(region_box, (mask_w, mask_h),
(width, height))
region_box = utils.generate_crop_region(region_box, (width, height))
region_box = np.vstack((region_box, np.array([0, 0, width - 1,
height - 1])))
gt_boxes, labels = dataset.get_gtbox(img_path)
chip_list, chip_gt_list, chip_label_list = generate_region_gt(
(width, height), region_box, gt_boxes, labels)
chip_loc = write_chip_and_anno(image, imgid, chip_list, chip_gt_list,
chip_label_list)
return len(chip_list), chip_loc
def _vis(img_path, dataset):
img = cv2.imread(img_path)
height, width = img.shape[:2]
img_id = os.path.basename(img_path)[:-4]
pred_mask_path = os.path.join(pred_mask_dir, img_id + '.png')
label_mask_path = os.path.join(segmentation_dir, img_id + '_region.png')
pred_mask = cv2.imread(pred_mask_path) * 255
label_mask = cv2.imread(label_mask_path) * 255
# bounding box
img1 = img.copy()
gt_box_list, _ = dataset.get_gtbox(img_path)
for box in gt_box_list:
cv2.rectangle(img1, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 4)
cv2.putText(img1, str((box[2], box[3])), (box[0], box[3]),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 4)
cv2.putText(img1, str((width, height)), (100, 100),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 0), 4)
label_region_box, _ = utils.generate_box_from_mask(label_mask[:, :, 0])
print('box', gt_box_list)
print('region', label_region_box)
label_region_box = utils.resize_box(label_region_box, (40, 30),
(width, height))
for box in label_region_box:
cv2.rectangle(img1, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 5)
# region box
img2 = img.copy()
mask_h, mask_w = pred_mask.shape[:2]
region_box, contours = utils.generate_box_from_mask(pred_mask[:, :, 0])
resize_region_box = utils.resize_box(region_box, (mask_w, mask_h),
(width, height))
for box in resize_region_box:
cv2.rectangle(img2, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 5)
# region postprocess
img3 = img.copy()
new_regions = utils.region_postprocess(region_box, contours,
(mask_w, mask_h))
resize_region_box = utils.resize_box(new_regions, (mask_w, mask_h),
(width, height))
# new_regions = utils.generate_crop_region(resize_region_box, (width, height))
for box in resize_region_box:
cv2.rectangle(img3, (box[0], box[1]), (box[2], box[3]), (255, 0, 0), 5)
img4 = img.copy()
# resize_region_box = utils.resize_box(temp, (mask_w, mask_h), (width, height))
new_regions = utils.generate_crop_region(resize_region_box,
(width, height))
for box in new_regions:
cv2.rectangle(img4, (box[0], box[1]), (box[2], box[3]), (255, 0, 0),
10)
plt.subplot(1, 1, 1)
plt.imshow(img1[:, :, [2, 1, 0]])
# plt.subplot(2, 3, 2); plt.imshow(img2[:, :, [2,1,0]])
# plt.subplot(2, 3, 3); plt.imshow(img3[:, :, [2,1,0]])
# plt.subplot(2, 3, 4); plt.imshow(label_mask[:, :, [2,1,0]])
# plt.subplot(2, 3, 5); plt.imshow(pred_mask[:, :, [2,1,0]])
# plt.subplot(2, 3, 6); plt.imshow(img4[:, :, [2,1,0]])
plt.show()
cv2.waitKey(0)
mask_object = []
undetected_img = []
pixel_num = []
for img_path in tqdm(val_list, ncols=80):
img_name = os.path.basename(img_path)
raw_file = os.path.join(mask_path, img_name[:-4]+'.png')
img = cv2.imread(img_path)
height, width = img.shape[:2]
mask_img = cv2.imread(raw_file, cv2.IMREAD_GRAYSCALE)
mask_h, mask_w = mask_img.shape[:2]
pixel_num.append(np.sum(mask_img))
label_box, _ = dataset.get_gtbox(img_path)
region_box, contours = utils.generate_box_from_mask(mask_img)
region_box = utils.region_postprocess(region_box, contours, (mask_w, mask_h))
region_box = utils.resize_box(region_box, (mask_w, mask_h), (width, height))
region_box = utils.generate_crop_region(region_box, (width, height))
count = 0
for box1 in label_box:
for box2 in region_box:
if utils.overlap(box2, box1):
count += 1
break
label_object.append(len(label_box))
detect_object.append(count)
mask_object.append(len(region_box))
if len(label_box) != count:
def _vis(img_path, dataset):
img = cv2.imread(img_path)
height, width = img.shape[:2]
img_id = os.path.basename(img_path)[:-4]
pred_mask_path = os.path.join(pred_mask_dir, img_id+'.png')
label_mask_path = os.path.join(segmentation_dir, img_id+'_region.png')
pred_mask = cv2.imread(pred_mask_path) * 255
label_mask = cv2.imread(label_mask_path) * 255
# bounding box
img1 = img.copy()
gt_box_list, _ = dataset.get_gtbox(img_path)
for box in gt_box_list:
cv2.rectangle(img1, (box[0], box[1]), (box[2], box[3]), (255, 127, 0), 3)
# cv2.putText(img1, str((box[2], box[3])), (box[0], box[3]), cv2.FONT_HERSHEY_SIMPLEX, 1, (0,0,0), 4)
# cv2.putText(img1, str((width, height)), (100, 100), cv2.FONT_HERSHEY_SIMPLEX, 1, (0,0,0), 4)
# label_region_box, _ = utils.generate_box_from_mask(label_mask[:, :, 0])
# label_region_box = utils.resize_box(label_region_box, (40, 30), (width, height))
# for box in label_region_box:
# cv2.rectangle(img1, (box[0], box[1]), (box[2], box[3]), (0, 127, 255), 5)
# region box
img2 = img.copy()
mask_h, mask_w = pred_mask.shape[:2]
region_box, contours = utils.generate_box_from_mask(pred_mask[:, :, 0])
resize_region_box = utils.resize_box(region_box, (mask_w, mask_h), (width, height))
for box in resize_region_box:
cv2.rectangle(img2, (box[0], box[1]), (box[2], box[3]), (0, 127, 255), 4)
# region postprocess
img3 = img.copy()
new_regions = utils.region_postprocess(region_box, contours, (mask_w, mask_h))
resize_region_box = utils.resize_box(new_regions, (mask_w, mask_h), (width, height))
# new_regions = utils.generate_crop_region(resize_region_box, (width, height))
for box in resize_region_box:
cv2.rectangle(img3, (box[0], box[1]), (box[2], box[3]), (0, 127, 255), 4)
img4 = img.copy()
# resize_region_box = utils.resize_box(temp, (mask_w, mask_h), (width, height))
new_regions = utils.generate_crop_region(resize_region_box, (width, height))
chip_list = []
for box in new_regions:
cv2.rectangle(img4, (box[0], box[1]), (box[2], box[3]), (0, 127, 255), 7)
chip_list.append(img[box[1]:box[3], box[0]:box[2], :].copy())
# heat map
img5 = img.copy().astype(np.float64)
pred_mask_path = os.path.join(pred_mask_dir, img_id+'_heat.png')
heat_mask = cv2.imread(pred_mask_path, cv2.IMREAD_GRAYSCALE)
heat_mask = cv2.resize(heat_mask, (width, height), interpolation=cv2.INTER_NEAREST)
img5[:, :, 2] += heat_mask / 255 * 150
plt.subplot(2, 3, 1); plt.imshow(img1[:, :, [2,1,0]])
plt.subplot(2, 3, 2); plt.imshow(img2[:, :, [2,1,0]])
plt.subplot(2, 3, 3); plt.imshow(img3[:, :, [2,1,0]])
plt.subplot(2, 3, 4); plt.imshow(label_mask[:, :, [2,1,0]])
plt.subplot(2, 3, 5); plt.imshow(pred_mask[:, :, [2,1,0]])
plt.subplot(2, 3, 6); plt.imshow(img4[:, :, [2,1,0]])
dirname = os.path.join(pred_mask_dir, os.path.pardir)
cv2.imwrite(os.path.join(dirname, 'virtualization', 'image.jpg'), img)
cv2.imwrite(os.path.join(dirname, 'virtualization', 'bbox.jpg'), img1)
cv2.imwrite(os.path.join(dirname, 'virtualization', 'brec.jpg'), img2)
cv2.imwrite(os.path.join(dirname, 'virtualization', 'post_process.jpg'), img3)
cv2.imwrite(os.path.join(dirname, 'virtualization', 'label_mask.jpg'), label_mask)
cv2.imwrite(os.path.join(dirname, 'virtualization', 'pred_mask.jpg'), pred_mask)
cv2.imwrite(os.path.join(dirname, 'virtualization', 'result.jpg'), img4)
cv2.imwrite(os.path.join(dirname, 'virtualization', 'heat.jpg'), img5)
for i, chip in enumerate(chip_list):
cv2.imwrite(os.path.join(dirname, 'virtualization', 'chip_%d.jpg' % i), chip)
plt.show()
cv2.waitKey(0)
