Retail Product Checkout Dataset Tools

1. Install

pip install rpctool

pip install git+https://github.com/DIYer22/retail_product_checkout_tools

2. Usage

python -m rpctool {result json} {ground truth json}

3. Help

$ python -m rpctool -h

Evaluate resFile with annFile and return evaluation result in markdown format

positional arguments:
  FILE        path to result json(support bbox format)
  FILE        path to ground truth json

optional arguments:
  -h, --help  show this help message and exit
  --mmap      Evaluate mAP50 and mmAP

4. Example

Input:

python -m rpctool bbox_results.json ~/retail_product_checkout/instances_test2019.json

Return:

## result on RPC-Dataset
|     diff |  method |   cAcc |  mCIoU |  ACD | mCCD |
|     ---: |    ---: |   ---: |   ---: | ---: | ---: |
|     easy | default | 63.19% | 90.64% | 0.72 | 0.11 |
|   medium | default | 43.02% | 90.64% | 1.24 | 0.11 |
|     hard | default | 31.01% | 90.41% | 1.77 |  0.1 |
| averaged | default |  45.6% | 90.58% | 1.25 |  0.1 |

5. Result Format

Result format (i.e bbox_results.json) should has the same data structure as the data struecture of COCO Object Detection Result Format :

[{
    "image_id"    : int, 
    "category_id" : int, 
    "bbox"        : [x,y,width,height], 
    "score"       : float,
}]