‘Results do not correspond to current coco set‘

In the training YOLOV6 When I met

'Results do not correspond to current coco set'

This problem .

assert set(annsImgIds) == (set(annsImgIds) & set(self.getImgIds()))  Check according to this assertion .

call coco API Check it out , Print out len((person_imgs_id) The value of is 297.

from pycocotools.coco import COCO
val_info = r'VOC/annotations/instances_val.json'

coco = COCO(val_info)  #  Import validation set 
person_imgs_id = coco.getImgIds()
print(len(person_imgs_id))

While checking set(annsImgIds) when , stay evaler.py Of documents eval_model In this method

annsImgIds = [ann['image_id'] for ann in pred_results]
print(len(set(annsImgIds)))

The printed length is 497（ That is, the number of my validation sets ）

And then I checked instances_val.json file , Find out

 image_id yes 6, And my data picture id by 6.8CAXX.jpg, Then we find the problem .

stay dataset.py Medium generate_coco_format_labels In the method

img_id = osp.basename(img_path).split('.')[0]

This causes the image to be divided by a point id,6.8CAXX.jpg Divided and multiplied 6、8CAXX、jpg Three strings , So it led to json In the document image_id by 6 The situation of .

terms of settlement ：

stay dataset.py Medium generate_coco_format_labels In the method

img_id = osp.basename(img_path).split('j')[0].strip('.')

take img_id Change to the above code , With j Division , Remove the remaining points , The code can successfully run through