Evaluate DETR model fine tuned on panoptic segmentation dataset

[lombardata]

Hi everybody, I was following the tutorial : https://colab.research.google.com/github/NielsRogge/Transformers-Tutorials/blob/master/DETR/Fine_tuning_DetrForSegmentation_on_custom_dataset_end_to_end_approach.ipynb for fine tune a detr model on a panoptic dataset and everything was good (thanks Niels) but when I come to the evaluation part I start having troubles. My first question is : Why we evaluate our model (trained on panoptic annotations) on a detection dataset (we define the CocoDetection class for evaluating while we train on the CocoPanoptic class)? # load ground truths base_ds = CocoDetection(img_folder='/home/mcontini/Desktop/PhD/detr_huggingface/coco/images/val2017', ann_file='/home/mcontini/Desktop/PhD/detr_huggingface/coco/images/panoptic_annotations_trainval2017/annotations/panoptic_val2017.json', feature_extractor=feature_extractor, train=False).coco Then I would like to follow the tutorial but did not found the evaluate json file : ann_file='/content/drive/MyDrive/DETR/COCO data/annotations/instances_val2017.json', on the coco website. Thank you in advance !