IDEA-Research / GroundingDINO

[ECCV 2024] Official implementation of the paper "Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection"
https://arxiv.org/abs/2303.05499
Apache License 2.0
6.83k stars 693 forks source link

Regarding the issue of model recognition errors #283

Open SunShineAwayD opened 10 months ago

SunShineAwayD commented 10 months ago

Here is the command I used and the results when using the pre-trained model on RefCOCO for REC recognition. However, the recognition for left and right directions is very poor.

python demo/image_demo.py test_images/two_human/ /home/wh/sjx/yuanban_mm/mmdetection/configs/grounding_dino/grounding_dino_swin-b_finetune_16xb2_1x_coco.py --weights /home/wh/sjx/yuanban_mm/mmdetection/weights/grounding_dino_swin-b_finetune_16xb2_1x_coco_20230921_153201-f219e0c0.pth --texts 'the left person.' --device 'cpu' --tokens-positive -1

1 000000000202 000000000890 000000003618