Releases: RangiLyu/nanodet
Releases · RangiLyu/nanodet
NanoDet-Plus v1.0.0
NanoDet-Plus v1.0.0
A stable version of NanoDet-Plus with PyTorch 1.x.
It requires pytorch-lightning>=1.9.0,<2.0.0 and torch>=1.10,<2.0.
What's Changed
- [Enhance]: Disable DDP when using a single GPU. by @RangiLyu in #369
- update version and readme by @RangiLyu in #370
- Update requirements.txt by @wwdok in #391
- [Feature]: Support print per class AP. by @RangiLyu in #395
- fix typo in README.md by @tuduweb in #397
- DEBUG: list type can not match cv2.mat by @Shawn-Tao in #423
- upgrade lightning version by @RangiLyu in #458
- Update README.md by @tpoisonooo in #438
- Fix demo libtorch error by @jedi007 in #420
- [Feature] Support timm backbones. by @RangiLyu in #399
- Add mem monitor and fix strategy by @RangiLyu in #462
- [Feature] Support tuning parameter-level optim hyperparameters. by @RangiLyu in #463
- [Feature] Support ignore boxes in nanodet head by @zero0kiriyu in #480
- Fixes a couple of issues to add fp16 training support by @RangiLyu in #488
- support lightning-1.9.0 by @RangiLyu in #489
- Implemented yolo dataset support by @cansik in #487
- bump version to 1.0.0 by @RangiLyu in #502
New Contributors
- @tuduweb made their first contribution in #397
- @Shawn-Tao made their first contribution in #423
- @tpoisonooo made their first contribution in #438
- @jedi007 made their first contribution in #420
- @zero0kiriyu made their first contribution in #480
- @cansik made their first contribution in #487
Full Changelog: v0.4.2...v1.0.0
Contributors
cansik, tpoisonooo, and 6 other contributors
Assets 2
NanoDet-Plus v1.0.0-alpha
NanoDet-Plus v1.0.0-alpha
In NanoDet-Plus, we propose a novel label assignment strategy with a simple assign guidance module (AGM) and a dynamic soft label assigner (DSLA) to solve the optimal label assignment problem in lightweight model training. We also introduce a light feature pyramid called Ghost-PAN to enhance multi-layer feature fusion. These improvements boost previous NanoDet's detection accuracy by 7 mAP on COCO dataset.
| Model | Resolution | mAPval 0.5:0.95 |
CPU Latency (i7-8700) |
ARM Latency (4xA76) |
FLOPS | Params | Model Size |
|---|---|---|---|---|---|---|---|
| NanoDet-m | 320*320 | 20.6 | 4.98ms | 10.23ms | 0.72G | 0.95M | 1.8MB(FP16) | 980KB(INT8) |
| NanoDet-Plus-m | 320*320 | 27.0 | 5.25ms | 11.97ms | 0.9G | 1.17M | 2.3MB(FP16) | 1.2MB(INT8) |
| NanoDet-Plus-m | 416*416 | 30.4 | 8.32ms | 19.77ms | 1.52G | 1.17M | 2.3MB(FP16) | 1.2MB(INT8) |
| NanoDet-Plus-m-1.5x | 320*320 | 29.9 | 7.21ms | 15.90ms | 1.75G | 2.44M | 4.7MB(FP16) | 2.3MB(INT8) |
| NanoDet-Plus-m-1.5x | 416*416 | 34.1 | 11.50ms | 25.49ms | 2.97G | 2.44M | 4.7MB(FP16) | 2.3MB(INT8) |
| YOLOv3-Tiny | 416*416 | 16.6 | - | 37.6ms | 5.62G | 8.86M | 33.7MB |
| YOLOv4-Tiny | 416*416 | 21.7 | - | 32.81ms | 6.96G | 6.06M | 23.0MB |
| YOLOX-Nano | 416*416 | 25.8 | - | 23.08ms | 1.08G | 0.91M | 1.8MB(FP16) |
| YOLOv5-n | 640*640 | 28.4 | - | 44.39ms | 4.5G | 1.9M | 3.8MB(FP16) |
| FBNetV5 | 320*640 | 30.4 | - | - | 1.8G | - | - |
| MobileDet | 320*320 | 25.6 | - | - | 0.9G | - | - |
Model checkpoints and weights
Download in the release files.
v0.4.2 Fix compatibility
v0.4.2
Fix some compatibility issue of NanoDet v0.4
Fix pytorch-lightning compatibility. (#304 #309 )
Fix pytorch1.9 compatibility. (#308 )
Support not raising an error when evaluate with empty results. (#310)
I'm doing a lot of refactoring. NanoDet v1.x is coming soon.
Download pretrained models
| Model | Backbone | Resolution | COCO mAP | FLOPS | Params | Pre-train weight | ncnn model | ncnn-int8 |
|---|---|---|---|---|---|---|---|---|
| NanoDet-m | ShuffleNetV2 1.0x | 320*320 | 20.6 | 0.72B | 0.95M | Download | Download | Download |
| NanoDet-m-416 | ShuffleNetV2 1.0x | 416*416 | 23.5 | 1.2B | 0.95M | Download | Download | Download |
| NanoDet-m-1.5x | ShuffleNetV2 1.5x | 320*320 | 23.5 | 1.44B | 2.08M | Download | Download | Download |
| NanoDet-m-1.5x-416 | ShuffleNetV2 1.5x | 416*416 | 26.8 | 2.42B | 2.08M | Download | Download | Download |
| NanoDet-t | ShuffleNetV2 1.0x | 320*320 | 21.7 | 0.96B | 1.36M | Download | ||
| NanoDet-g | Custom CSP Net | 416*416 | 22.9 | 4.2B | 3.81M | Download | ||
| NanoDet-EfficientLite | EfficientNet-Lite0 | 320*320 | 24.7 | 1.72B | 3.11M | Download | ||
| NanoDet-EfficientLite | EfficientNet-Lite1 | 416*416 | 30.3 | 4.06B | 4.01M | Download | ||
| NanoDet-EfficientLite | EfficientNet-Lite2 | 512*512 | 32.6 | 7.12B | 4.71M | Download | ||
| NanoDet-RepVGG | RepVGG-A0 | 416*416 | 27.8 | 11.3B | 6.75M | Download |
v0.4.1
v0.4.1
This is a final release of NanoDet v0.x.
I'm doing a lot of refactoring. NanoDet v1.x is coming soon.
Download pretrained models
| Model | Backbone | Resolution | COCO mAP | FLOPS | Params | Pre-train weight | ncnn model | ncnn-int8 |
|---|---|---|---|---|---|---|---|---|
| NanoDet-m | ShuffleNetV2 1.0x | 320*320 | 20.6 | 0.72B | 0.95M | Download | Download | Download |
| NanoDet-m-416 | ShuffleNetV2 1.0x | 416*416 | 23.5 | 1.2B | 0.95M | Download | Download | Download |
| NanoDet-m-1.5x | ShuffleNetV2 1.5x | 320*320 | 23.5 | 1.44B | 2.08M | Download | Download | Download |
| NanoDet-m-1.5x-416 | ShuffleNetV2 1.5x | 416*416 | 26.8 | 2.42B | 2.08M | Download | Download | Download |
| NanoDet-t | ShuffleNetV2 1.0x | 320*320 | 21.7 | 0.96B | 1.36M | Download | ||
| NanoDet-g | Custom CSP Net | 416*416 | 22.9 | 4.2B | 3.81M | Download | ||
| NanoDet-EfficientLite | EfficientNet-Lite0 | 320*320 | 24.7 | 1.72B | 3.11M | Download | ||
| NanoDet-EfficientLite | EfficientNet-Lite1 | 416*416 | 30.3 | 4.06B | 4.01M | Download | ||
| NanoDet-EfficientLite | EfficientNet-Lite2 | 512*512 | 32.6 | 7.12B | 4.71M | Download | ||
| NanoDet-RepVGG | RepVGG-A0 | 416*416 | 27.8 | 11.3B | 6.75M | Download |
v0.4.0
What's new in v0.4.0
- Fix a little bug in demo.py by BlainWu (#210)
- Add script to export TorchScript model by strawberrypie (#211)
- Use fixed output names when exporting ONNX (#218)
- Use scale_factor instead of fixed size in resize to support dynamic shape inference (#218)
- Ensure num_classes equal len(class_names) by ZHEQIUSHUI (#221)
- Fix a bug in mnn demo while using GPU device by AcherStyx (#234)
- Fix with_last_conv bug in shufflenet (#239)
- Support batch eval (#241)
- Add nanodet-m-1.5x models (#242)
- Update model benchmark (#246)
- Prevent lightning Trainer from disabling cudnn.benchmark (#249)
- Fix multi-GPU evaluation bug with pytorch-lightning (#254)
Download pretrained models
| Model | Backbone | Resolution | COCO mAP | FLOPS | Params | Pre-train weight |
|---|---|---|---|---|---|---|
| NanoDet-m | ShuffleNetV2 1.0x | 320*320 | 20.6 | 0.72B | 0.95M | Download |
| NanoDet-m-416 | ShuffleNetV2 1.0x | 416*416 | 23.5 | 1.2B | 0.95M | Download |
| NanoDet-m-1.5x | ShuffleNetV2 1.5x | 320*320 | 23.5 | 1.44B | 2.08M | Download |
| NanoDet-m-1.5x-416 | ShuffleNetV2 1.5x | 416*416 | 26.8 | 2.42B | 2.08M | Download |
| NanoDet-t | ShuffleNetV2 1.0x | 320*320 | 21.7 | 0.96B | 1.36M | Download |
| NanoDet-g | Custom CSP Net | 416*416 | 22.9 | 4.2B | 3.81M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite0 | 320*320 | 24.7 | 1.72B | 3.11M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite1 | 416*416 | 30.3 | 4.06B | 4.01M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite2 | 512*512 | 32.6 | 7.12B | 4.71M | Download |
| NanoDet-RepVGG | RepVGG-A0 | 416*416 | 27.8 | 11.3B | 6.75M | Download |
Download ncnn models below
v0.3.0 Release
What's new in v0.3.0
- Refactor training and testing code with pytorch-lightning.
- Solving ONNX inference AxisError by zshn25 (#198).
Download pretrained models
| Model | Backbone | Resolution | COCO mAP | FLOPS | Params | Pre-train weight |
|---|---|---|---|---|---|---|
| NanoDet-m | ShuffleNetV2 1.0x | 320*320 | 20.6 | 0.72B | 0.95M | Download |
| NanoDet-m-416 | ShuffleNetV2 1.0x | 416*416 | 23.5 | 1.2B | 0.95M | Download |
| NanoDet-t (NEW) | ShuffleNetV2 1.0x | 320*320 | 21.7 | 0.96B | 1.36M | Download |
| NanoDet-g | Custom CSP Net | 416*416 | 22.9 | 4.2B | 3.81M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite0 | 320*320 | 24.7 | 1.72B | 3.11M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite1 | 416*416 | 30.3 | 4.06B | 4.01M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite2 | 512*512 | 32.6 | 7.12B | 4.71M | Download |
| NanoDet-RepVGG | RepVGG-A0 | 416*416 | 27.8 | 11.3B | 6.75M | Download |
v0.2.0 Release
What's new in v0.2.0
- Add pyncnn demo by caishanli (#167).
- Fix ncnn demo build failure without vulkan by nihui (#168).
- Add NanoDet-t with Transformer Attention Network (#183).
- Add Notebook demo by zhiqwang (#188).
- Add feature of saving demo inference result by wwdok (#191).
- Fix utf-8 decode bug (#184).
- Fix test bug.
Download pretrained models
| Model | Backbone | Resolution | COCO mAP | FLOPS | Params | Pre-train weight |
|---|---|---|---|---|---|---|
| NanoDet-m | ShuffleNetV2 1.0x | 320*320 | 20.6 | 0.72B | 0.95M | Download |
| NanoDet-m-416 | ShuffleNetV2 1.0x | 416*416 | 23.5 | 1.2B | 0.95M | Download |
| NanoDet-t (NEW) | ShuffleNetV2 1.0x | 320*320 | 21.7 | 0.96B | 1.36M | Download |
| NanoDet-g | Custom CSP Net | 416*416 | 22.9 | 4.2B | 3.81M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite0 | 320*320 | 24.7 | 1.72B | 3.11M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite1 | 416*416 | 30.3 | 4.06B | 4.01M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite2 | 512*512 | 32.6 | 7.12B | 4.71M | Download |
| NanoDet-RepVGG | RepVGG-A0 | 416*416 | 27.8 | 11.3B | 6.75M | Download |
v0.1.0 Release
What's new in v0.1.0
- Support MNN python and cpp inference (#83 ).
- Support OpenVINO inference.
- Support libtorch inference experimentally.
- Add NanoDet-g.
- Add EfficientNet-Lite and Rep-VGG backbone.
- Add Model Zoo and provide more pre-trained model.
- Refactor GFL head (#154 ).
Download pretrained models
| Model | Backbone | Resolution | COCO mAP | FLOPS | Params | Pre-train weight |
|---|---|---|---|---|---|---|
| NanoDet-m | ShuffleNetV2 1.0x | 320*320 | 20.6 | 0.72B | 0.95M | Download |
| NanoDet-m-416 | ShuffleNetV2 1.0x | 416*416 | 23.5 | 1.2B | 0.95M | Download |
| NanoDet-g | Custom CSP Net | 416*416 | 22.9 | 4.2B | 3.81M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite0 | 320*320 | 24.7 | 1.72B | 3.11M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite1 | 416*416 | 30.3 | 4.06B | 4.01M | Download |
| NanoDet-EfficientLite | EfficientNet-Lite2 | 512*512 | 32.6 | 7.12B | 4.71M | Download |
| NanoDet-RepVGG | RepVGG-A0 | 416*416 | 27.8 | 11.3B | 6.75M | Download |