Convnext_alpha/object_detection

• COCO Object detection with ConvNeXt
• • Getting started
  • Results and Fine-tuned Models
  • • Training
    • Inference
  • Acknowledgment

COCO Object detection with ConvNeXt

Getting started

We add ConvNeXt model and config files to Swin Detection.
Our code has been tested with commit 6a979e2. Please refer to README.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
ConvNeXt-T	ImageNet-1K	Mask R-CNN	3x	46.2	41.7	48M	262G	model
ConvNeXt-T	ImageNet-1K	Cascade Mask R-CNN	3x	50.4	43.7	86M	741G	model
ConvNeXt-S	ImageNet-1K	Cascade Mask R-CNN	3x	51.9	45.0	108M	827G	model
ConvNeXt-B	ImageNet-1K	Cascade Mask R-CNN	3x	52.7	45.6	146M	964G	model
ConvNeXt-B	ImageNet-22K	Cascade Mask R-CNN	3x	54.0	46.9	146M	964G	model
ConvNeXt-L	ImageNet-22K	Cascade Mask R-CNN	3x	54.8	47.6	255M	1354G	model
ConvNeXt-XL	ImageNet-22K	Cascade Mask R-CNN	3x	55.2	47.7	407M	1898G	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 ConvNeXt-T backbone and 8 gpus, run:

tools/dist_train.sh configs/convnext/cascade_mask_rcnn_convnext_tiny_patch4_window7_mstrain_480-800_giou_4conv1f_adamw_3x_coco_in1k.py 8 --cfg-options model.pretrained=https://dl.fbaipublicfiles.com/convnext/convnext_tiny_1k_224.pth

More config files can be found at configs/convnext.

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.