IDEA-Research / DINO

[ICLR 2023] Official implementation of the paper "DINO: DETR with Improved DeNoising Anchor Boxes for End-to-End Object Detection"
Apache License 2.0
2.08k stars 228 forks source link

Inference scale on DINO model #242

Open truong2710-cyber opened 6 months ago

truong2710-cyber commented 6 months ago

Hi everyone, I have a trained DINO-5scale model on my custom dataset and I want to infer on it.

But when I upscale all of the images to 2x, I notice that the AP score drop heavily. My dataset contains mainly people and vehicles from a camera. In contrast, I did the same upscaling with YOLOv5 or Faster-RCNN and observed the higher AP score when upscaling. I think that it is due to the nature of Transformer-based model, but I'm not sure about that. So why that happens? Thank you.

pof1423 commented 5 months ago

i also experienced same! when i down scaling with cv2, the map is almost close to 0. someone explain it please !!

truong2710-cyber commented 2 months ago

has anyone known the reason yet :(((