coco test-dev

[helq2612]

Hi,

Great work! Do you have the code for coco test-dev?

Thanks!

[tangjiuqi097]

@helq2612 Hi,

You can modify the path to the test-dev.

In order to save the results, you should add some lines as follows:

# in datasets/coco_eval.py
class CocoEvaluator(object):
    def __init__(self, coco_gt, iou_types):
        assert isinstance(iou_types, (list, tuple))
        coco_gt = copy.deepcopy(coco_gt)
        self.coco_gt = coco_gt

        self.iou_types = iou_types
        self.coco_eval = {}
        for iou_type in iou_types:
            self.coco_eval[iou_type] = COCOeval(coco_gt, iouType=iou_type)

        self.img_ids = []
        self.eval_imgs = {k: [] for k in iou_types}
        self.results=[]

    def update(self, predictions):
        img_ids = list(np.unique(list(predictions.keys())))
        self.img_ids.extend(img_ids)

        for iou_type in self.iou_types:
            results = self.prepare(predictions, iou_type)
            self.results.extend(results)

# in main.py
            with open(os.path.join(args.output_dir,'res.json'),'w') as f:
                json.dump(coco_evaluator.results,f)

BTW, if you want to use multiple gpus, you should gather the results at each card.

[helq2612]

Thank you very much! I find 1 GPU works for me, but with multiple GPUs, the results seem not good.

[tangjiuqi097]

Yes, only the results at 1 card will be saved if simply using multiple GPUs. You can use the all_gather function to gather all the results at different ranks if you want to use multiple GPUs.

[tangjiuqi097]

Now we support to save the json results. And the results at different ranks will be gathered if you use multiple gpus.