COCO Object detection with RevCol
Getting started
We build RevCol object detection model based on mmdetection commit 3e26931. We add RevCol model and config files to the original repo. Please refer to get_started.md for installation and dataset preparation instructions.
Results and Fine-tuned Models
| name | Pretrained Model | Method | Lr Schd | box mAP | mask mAP | #params | FLOPs | Fine-tuned Model |
|---|---|---|---|---|---|---|---|---|
| RevCol-T | ImageNet-1K | Cascade Mask R-CNN | 3x | 50.6 | 43.8 | 88M | 741G | model |
| RevCol-S | ImageNet-1K | Cascade Mask R-CNN | 3x | 52.6 | 45.5 | 118M | 833G | model |
| RevCol-B | ImageNet-1K | Cascade Mask R-CNN | 3x | 53.0 | 45.9 | 196M | 988G | model |
| RevCol-B | ImageNet-22K | Cascade Mask R-CNN | 3x | 55.0 | 47.5 | 196M | 988G | model |
| RevCol-L | ImageNet-22K | Cascade Mask R-CNN | 3x | 55.9 | 48.4 | 330M | 1453G | model |
Training
To train a detector with pre-trained models, run:
# single-gpu training
python tools/train.py <CONFIG_FILE> --cfg-options model.pretrained=<PRETRAIN_MODEL> [other optional arguments]
# multi-gpu training
tools/dist_train.sh <CONFIG_FILE> <GPU_NUM> --cfg-options model.pretrained=<PRETRAIN_MODEL> [other optional arguments]
For example, to train a Cascade Mask R-CNN model with a RevCol-T backbone and 8 gpus, run:
tools/dist_train.sh configs/revcol/cascade_mask_rcnn_revcol_tiny_3x_in1k.py 8 --cfg-options pretrained=<PRETRAIN_MODEL>
More config files can be found at configs/revcol.
Inference
# single-gpu testing
python tools/test.py <CONFIG_FILE> <DET_CHECKPOINT_FILE> --eval bbox segm
# multi-gpu testing
tools/dist_test.sh <CONFIG_FILE> <DET_CHECKPOINT_FILE> <GPU_NUM> --eval bbox segm
Acknowledgment
This code is built using mmdetection, timm libraries, and BeiT, Swin Transformer repositories.