Results format
Result的格式与ground truth的annotation format形式类似。所有的结果都被存成一个JSON文件(save via gason in Matlab or json.dump in Python).
可参考示例结果的JSON文件 coco/results/
用于可视化ground truth的一些官方cocoapi也可用于这些结果文件,For details please see evalDemo (demo) and also loadRes() in the CocoApi
结果的数据结构如下(共bounding box, segmentation ,keypoint detection, stuff segmentation,caption generation五种,后四种暂时忽略):
注:Box的坐标为浮点数,以左上角为起点(包含0),建议四舍五入坐标到0.1精度来减小JSON文件大小
Upload Results to Evaluation Server
可以在val上跑coco本地的API的evaluation code得到结果得到结果,然后提交到评估服务器得到在test上的结果。
(1) Create an account on CodaLab. This will allow you to participate in all COCO challenges.
(2) Carefully review the guidelines for entering the COCO challenges and using the test sets.
(3) Prepare a JSON file containing your results in the correct results format for the challenge you wish to enter.
(4) File naming: the JSON file should be named "[type]_[testset]_[alg]_results.json". Replace [type] with the challenge type, [testset] with the test split you are using, and [alg] with your algorithm name. Finally place the JSON file into a zip file named "[type]_[testset]_[alg]_results.zip". Summary:
(5) To submit your zipped result file to the COCO Challenge click on the “Participate” tab on the appropriate CodaLab evaluation server. Select the test split and for detection only the test type (bbox or segm). When you select “Submit / View Results” you will be given the option to submit new results. Please fill in the required fields and click “Submit”. A pop-up will prompt you to select the results zip file for upload. After the file is uploaded the evaluation server will begin p