Generate a tiny coco dataset for training debug. The training and test sets each contain 50 images and the corresponding instance, keypoint, and capture tags. The dataset file structure as follows:
tiny_coco
|-annotations
| |-instances_train2017.json
| |-instances_val2017.json
| |-...
|
|-train2017
| |-000000005802.jpg
| |-000000005803.jpg
| |-...
|
|-val2017
| |-000000005802.jpg
| |-000000005803.jpg
| |-...
Reference from zhihu. You can re-slice the data size through the original coco dataset(18G) or the current tiny coco dataset