- ubuntu 18.04
- pytorch
- tiny PASCAL VOC dataset
- contains only 1,349 training images, 100 test images with 20 common object classes
-
first modify the
data/config.py. Create a definition for your dataset. if you don't have validation data, use the same path to training_inages for path_to_validation_images.In
train.py--parser.add_argument('--validation_size', default=200, type=int, help='The number of images to use for validation.')will use the first 200 images for validation.
my_custom_dataset = dataset_base.copy({
'name': 'My Dataset',
'train_images': 'path_to_training_images',
'train_info': 'path_to_training_annotation',
'valid_images': 'path_to_validation_images',
'valid_info': 'path_to_validation_annotation',
'has_gt': True,
'class_names': ('my_class_id_1', 'my_class_id_2', 'my_class_id_3', ...)
})
- then turn
yolact_base_config = 'dataset'toyolact_base_config = 'my_custom_dataset'
- add
--parser.add_argument('--config', defalut=yolact_base_config) - you can choose learning rate for SGD, batchsize, epoch...
- run
python3 train.pyto train your model.
-
run
python3 eval.pyto get image with mask and bounding box on detected item. -
you can choose weight, score_threshold, number of item.
-
add the test images to
data/test_imagesand output will indata/output -
example:
-
you will get output images and one json file.
- run
python3 tococojson.pyit can turn the original json file into coco style format.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent