Skip to content

Latest commit

 

History

History
216 lines (188 loc) · 6.83 KB

README.md

File metadata and controls

216 lines (188 loc) · 6.83 KB

NAT - Object Detection

Requirements

Python 3.7 is strongly encouraged. For ease of use, you can just set up a new environment and run the following:

pip3 install -r requirements.txt

Similar to and because of classification, object detection also depends on timm, ninja, and fvcore. Object detection experiments were conducted with mmdetection. The following are the recommended versions of these libraries and are strongly encouraged for reproducibility and speed:

torch==1.11.0+cu113
torchvision==0.12.0+cu113
timm==0.5.0
ninja==1.10.2.3
fvcore==0.1.5.post20220305

mmcv-full==1.4.8
mmdet==2.19.0

Setup

The kernel is identical to the one in classification, it is just configured differently for downstream tasks because they are more resolution-heavy, as opposed to the former which is batch-heavy.

Once you have the requirements, please run the following and allow a few minutes for Ninja to compile the extension:

python3 cuda/gradcheck.py

Ninja is not verbose while compiling, so it is normal not to see any outputs unless there is an error. After a few minutes you should see the message "Verifying backward pass...". You should see two tests pass afterwards to verify that the extension is functioning normally. Once it is done, you can start training normally.

It is recommended to repeat this step upon every update to the repository.

You can find more details about the extension here.

Training on COCO

Mask R-CNN

NAT-Mini + Mask R-CNN 
./dist_train.sh configs/nat/mask_rcnn_nat_mini_3x_coco.py $NUM_GPUS --cfg-options data.samples_per_gpu=$((16/$NUM_GPUS)) data.workers_per_gpu=$((16/$NUM_GPUS))
NAT-Tiny + Mask R-CNN 
./dist_train.sh configs/nat/mask_rcnn_nat_tiny_3x_coco.py $NUM_GPUS --cfg-options data.samples_per_gpu=$((16/$NUM_GPUS)) data.workers_per_gpu=$((16/$NUM_GPUS))
NAT-Small + Mask R-CNN 
./dist_train.sh configs/nat/mask_rcnn_nat_small_3x_coco.py $NUM_GPUS --cfg-options data.samples_per_gpu=$((16/$NUM_GPUS)) data.workers_per_gpu=$((16/$NUM_GPUS))

Cascade Mask R-CNN

NAT-Mini + Cascade R-CNN 
./dist_train.sh configs/nat/cascade_mask_rcnn_nat_mini_3x_coco.py $NUM_GPUS --cfg-options data.samples_per_gpu=$((16/$NUM_GPUS)) data.workers_per_gpu=$((16/$NUM_GPUS))
NAT-Tiny + Mask R-CNN 
./dist_train.sh configs/nat/cascade_mask_rcnn_nat_tiny_3x_coco.py $NUM_GPUS --cfg-options data.samples_per_gpu=$((16/$NUM_GPUS)) data.workers_per_gpu=$((16/$NUM_GPUS))
NAT-Small + Mask R-CNN 
./dist_train.sh configs/nat/cascade_mask_rcnn_nat_small_3x_coco.py $NUM_GPUS --cfg-options data.samples_per_gpu=$((16/$NUM_GPUS)) data.workers_per_gpu=$((16/$NUM_GPUS))
NAT-Base + Cascade Mask R-CNN 
./dist_train.sh configs/nat/cascade_mask_rcnn_nat_base_3x_coco.py $NUM_GPUS --cfg-options data.samples_per_gpu=$((16/$NUM_GPUS)) data.workers_per_gpu=$((16/$NUM_GPUS))

Validation

Mask R-CNN

NAT-Mini + Mask R-CNN 
./dist_test.sh \
    configs/nat/mask_rcnn_nat_mini_3x_coco.py \
    http://ix.cs.uoregon.edu/~alih/nat/checkpoints/DET/nat_mini_maskrcnn.pth \
    $NUM_GPUS \
    --eval bbox segm
NAT-Tiny + Mask R-CNN 
./dist_test.sh \
    configs/nat/mask_rcnn_nat_tiny_3x_coco.py \
    http://ix.cs.uoregon.edu/~alih/nat/checkpoints/DET/nat_tiny_maskrcnn.pth \
    $NUM_GPUS \
    --eval bbox segm
NAT-Small + Mask R-CNN 
./dist_test.sh \
    configs/nat/mask_rcnn_nat_small_3x_coco.py \
    http://ix.cs.uoregon.edu/~alih/nat/checkpoints/DET/nat_small_maskrcnn.pth \
    $NUM_GPUS \
    --eval bbox segm

Cascade Mask R-CNN

NAT-Mini + Cascade R-CNN 
./dist_test.sh \
    configs/nat/cascade_mask_rcnn_nat_mini_3x_coco.py \
    http://ix.cs.uoregon.edu/~alih/nat/checkpoints/DET/nat_mini_cascademaskrcnn.pth \
    $NUM_GPUS \
    --eval bbox segm
NAT-Tiny + Cascade Mask R-CNN 
./dist_test.sh \
    configs/nat/cascade_mask_rcnn_nat_tiny_3x_coco.py \
    http://ix.cs.uoregon.edu/~alih/nat/checkpoints/DET/nat_tiny_cascademaskrcnn.pth \
    $NUM_GPUS \
    --eval bbox segm
NAT-Small + Cascade Mask R-CNN 
./dist_test.sh \
    configs/nat/cascade_mask_rcnn_nat_small_3x_coco.py \
    http://ix.cs.uoregon.edu/~alih/nat/checkpoints/DET/nat_small_cascademaskrcnn.pth \
    $NUM_GPUS \
    --eval bbox segm
NAT-Base + Cascade Mask R-CNN 
./dist_test.sh \
    configs/nat/cascade_mask_rcnn_nat_base_3x_coco.py \
    http://ix.cs.uoregon.edu/~alih/nat/checkpoints/DET/nat_base_cascademaskrcnn.pth \
    $NUM_GPUS \
    --eval bbox segm

Checkpoints

Backbone Network # of Params FLOPs mAP Mask mAP Checkpoint Config
NAT-Mini Mask R-CNN 40M 225G 46.5 41.7 Download config.py
NAT-Tiny Mask R-CNN 48M 258G 47.7 42.6 Download config.py
NAT-Small Mask R-CNN 70M 330G 48.4 43.2 Download config.py
NAT-Mini Cascade Mask R-CNN 77M 704G 50.3 43.6 Download config.py
NAT-Tiny Cascade Mask R-CNN 85M 737G 51.4 44.5 Download config.py
NAT-Small Cascade Mask R-CNN 108M 809G 52.0 44.9 Download config.py
NAT-Base Cascade Mask R-CNN 147M 931G 52.3 45.1 Download config.py