RTDETRv2 re-implementation using PyTorch
git clone https://github.com/balalofernandez/RTDETRv2-pt.git
cd RTDETRv2-pt
conda create -n rtdetr python=3.10
conda activate rtdetr
pip install -r requirements.txt
- Use your own dataset, returning similar dictionary to the one used in
/RTDETRv2-pt/data/coco_dataset.py - Adapt the number of classes in the yml file
[num_classes]to your number of classes + 1 (90 + 1 = 91 in COCO) - Modify
train.pyto include your dataset.
This version uses pretrained versions of Resnet for the backbone. However, base models from the original repo should work but don't forget to modify the config file and update the path to the checkpoint in args.
| Model | Dataset | Input Size | APval | AP50val | #Params(M) | FPS | config | checkpoint |
|---|---|---|---|---|---|---|---|---|
| RT-DETRv2-S | COCO | 640 | 48.1 (+1.6) | 65.1 | 20 | 217 | config | url |
| RT-DETRv2-M | COCO | 640 | 49.9 (+1.0) | 67.5 | 31 | 161 | config | url |
| RT-DETRv2-M | COCO | 640 | 51.9 (+0.6) | 69.9 | 36 | 145 | config | url |
| RT-DETRv2-L* | COCO | 640 | 53.4 (+0.3) | 71.6 | 42 | 108 | config | url |
| RT-DETRv2-X | COCO | 640 | 54.3 | 72.8 (+0.1) | 76 | 74 | config | url |
*is the current configuration.
