개정판 1e8ea226
issue #1366: test big symbol
Change-Id: I42759e4e53529fc18b1fcb71e2a31ba90ea81f6d
| DTI_PID/WebServer/app/training/index.py | ||
|---|---|---|
| 84 | 84 |
data_path = os.path.join(os.path.dirname(os.path.realpath(__file__)) + '\\..\\..\\symbol_training\\Data\\', datas['name']) |
| 85 | 85 | |
| 86 | 86 |
if os.path.isdir(data_path): |
| 87 |
train.train(name=datas['name'], classes=datas['classes'], root_path=data_path, pre_trained_model_path=os.path.dirname(os.path.realpath( |
|
| 87 |
train.train(name=datas['name'], classes=datas['classes'], bigs=datas['bigs'], root_path=data_path, pre_trained_model_path=os.path.dirname(os.path.realpath(
|
|
| 88 | 88 |
__file__)) + '\\..\\..\\symbol_training\\pre_trained_model\\only_params_trained_yolo_voc') |
| 89 | 89 |
|
| 90 | 90 |
return jsonify({'count': 1})
|
| DTI_PID/WebServer/symbol_recognition/test_doftech_all_images.py | ||
|---|---|---|
| 64 | 64 |
with open(os.path.join(saved_path, name + "_info.info"), 'r') as stream: |
| 65 | 65 |
con = stream.read().split('\n')
|
| 66 | 66 |
classes = int(con[0]) |
| 67 |
DOFTECH_CLASSES = con[1:] |
|
| 67 |
DOFTECH_CLASSES = con[1:con.index('***bigs***')]
|
|
| 68 |
bigs = con[con.index('***bigs***') + 1:]
|
|
| 68 | 69 | |
| 69 | 70 |
if torch.cuda.is_available(): |
| 70 | 71 |
model1 = Yolo(classes).cuda() |
| ... | ... | |
| 147 | 148 |
print("=========> SMALL OBJECT DETECTION")
|
| 148 | 149 |
for idx_small in range(len(small_object_patch_list)): |
| 149 | 150 |
patchs = small_object_patch_list[idx_small] |
| 150 |
total_symbole_list.append(detection_object(patchs, model1, opc=False, opt=opt, save_root=save_dir_tile1)) |
|
| 151 |
total_symbole_list.append(detection_object(patchs, model1, opc=False, opt=opt, save_root=save_dir_tile1, bigs=bigs))
|
|
| 151 | 152 |
# ---------------------------------------------- |
| 152 | 153 |
# Large Object Detection (OPC etc.) |
| 153 | 154 |
# ---------------------------------------------- |
| 154 | 155 |
print("=========> LARGE OBJECT DETECTION")
|
| 155 | 156 |
for idx_large in range(len(large_object_patch_list)): |
| 156 | 157 |
patchs = large_object_patch_list[idx_large] |
| 157 |
total_symbole_list.append(detection_object(patchs, model2, opc=True, opt=opt, save_root=save_dir_tile2)) |
|
| 158 |
total_symbole_list.append(detection_object(patchs, model2, opc=True, opt=opt, save_root=save_dir_tile2, bigs=bigs))
|
|
| 158 | 159 | |
| 159 | 160 |
t_image = img_list[idx_img].copy() |
| 160 | 161 |
count_result = merge_fn(t_image, total_symbole_list, save_dir) |
| ... | ... | |
| 175 | 176 |
return total_symbole_lists |
| 176 | 177 | |
| 177 | 178 | |
| 178 |
def detection_object(patchs, model, opc, opt, save_root=None): |
|
| 179 |
def detection_object(patchs, model, opc, opt, save_root=None, bigs=None):
|
|
| 179 | 180 |
global DOFTECH_CLASSES |
| 180 | 181 |
|
| 181 | 182 |
symbol_list = [] |
| ... | ... | |
| 204 | 205 |
predictions = predictions[0] |
| 205 | 206 |
output_image = cv2.cvtColor(np.array(pil_image.img), cv2.COLOR_RGB2BGR) |
| 206 | 207 |
for pred in predictions: |
| 207 |
if opc == True : |
|
| 208 |
if opc == True and pred[5] in bigs:
|
|
| 208 | 209 |
if pred[4] > 0.4:#pred[5] == "opc" and pred[4] > 0.4: # Classification threshold |
| 209 | 210 |
xmin = int(max(pred[0] / width_ratio, 0)) |
| 210 | 211 |
ymin = int(max(pred[1] / height_ratio, 0)) |
| ... | ... | |
| 224 | 225 |
output_image, pred[5] + ' : %.2f' % pred[4], |
| 225 | 226 |
(xmin, ymin + text_size[1] + 4), cv2.FONT_HERSHEY_PLAIN, 1, |
| 226 | 227 |
(255, 255, 255), 1) |
| 227 |
else :
|
|
| 228 |
elif opc == False and pred[5] not in bigs:
|
|
| 228 | 229 |
if pred[4] > 0.1: # Classification threshold |
| 229 | 230 |
xmin = int(max(pred[0] / width_ratio, 0)) |
| 230 | 231 |
ymin = int(max(pred[1] / height_ratio, 0)) |
| DTI_PID/WebServer/symbol_training/train.py | ||
|---|---|---|
| 31 | 31 | |
| 32 | 32 |
#print(len(DOFTECH_CLASSES)) |
| 33 | 33 | |
| 34 |
def train(name=None, classes=None, root_path=None, pre_trained_model_path=None): |
|
| 34 |
def train(name=None, classes=None, bigs=None, root_path=None, pre_trained_model_path=None):
|
|
| 35 | 35 |
global DOFTECH_CLASSES |
| 36 | 36 |
DOFTECH_CLASSES = classes |
| 37 | 37 | |
| ... | ... | |
| 72 | 72 |
with open(os.path.join(opt.saved_path, name + "_info.info"), 'w') as stream: |
| 73 | 73 |
con = str(len(DOFTECH_CLASSES)) |
| 74 | 74 |
names = '\n'.join(DOFTECH_CLASSES) |
| 75 |
con = con + '\n' + names |
|
| 75 |
bigs = '\n'.join(bigs) |
|
| 76 |
con = con + '\n' + names + '\n' + '***bigs***' + '\n' + bigs |
|
| 76 | 77 |
stream.write(con) |
| 77 | 78 | |
| 78 | 79 |
if torch.cuda.is_available(): |
내보내기 Unified diff