Queries regarding my generated Test_info and Human_detection JSON files

[Dipankar1997161]

Hello, Thank you for posting your work! @mks0601

I have gone through the Issues tab and tried to figure out what exactly this human_detection.json is.

From my perspective, human_detection is just a JSON file with keys like image_id, bbox, score, and category_id. (in short, segmented image properties)

Keys like filename, id, and image shape(H, W) are present in COCO image_info_test_dev.json file and image_info_test.json file but no score or BBOX nor keypoints
Is this correct or anything you wish to add?

Also, it is mentioned in the README.md that human_detection.json is used during testing stages, then what is the use of the COCO test JSON file and images which is under the test and annotation folders respectively??

I am asking this so I can properly generate those files for my custom dataset.
Because, I have my own validation set, Train set and Test set ready as per coco.

The only confusion that arises is how your human_detection.json looks! Is it as per the details I mentioned?
If possible, could you just share a snippet of the human_detection.json file that YOU generated? Visualizing that I can generate mine!!

I share a snippet of my human_detection.json file and image info . Please correct me here
Please note! I am trying to train on my custom dataset. so the test image info and the human detection json are for the same images. So is it fine to have for same images or do I need different ones?

1. Human_detection.json
   

2. Image_info_test_dev.json

Would like to hear from you

[mks0601]

human_detection.json alone is not enough as file_name is not provided in that file. The human_detection.json is not made by myself, but just an output of existing human detectin networks. human_detection.json has the required format for the COCO object detection evaluation. You can visit COCO dataset homepage.

[Dipankar1997161]

Actually, I have a JSON file (person_keypoints_scenario) that includes keys like file name, image shape, ID, bbox, and also key points. From this file, I omitted the file name and the keypoints to get the human_detection.json

But you are saying that's not enough as we do need a "file_name!"

Also, human_detection.json is used for testing after my model is created!

PLease note:
My university has asked me to test my model on the above mentioned JSON file (person_keypoints)

SO if I just replace the human_detection with the person_keypoint json file and make the change in the config.py
Won't that work??