/image_classification_pytorch

Primary LanguagePython

ImageNet Classification

Requirements

  • We recommend you to use Anaconda to create a conda environment:
conda create -n imagenet python=3.6
  • Then, activate the environment:
conda activate imagenet
  • Requirements:
pip install -r requirements.txt

PyTorch >= 1.9.1 and Torchvision >= 0.10.1

Experiments

ImageNet val

  • YOLOv2~v4's Backbone
Model Epoch size acc@1 GFLOPs Params Weight
DarkNet-19 90 224 72.9 5.4 20.8 M ckpt
DarkNet-53-SiLU 100 224 74.4 14.3 41.6 M ckpt
CSP-DarkNet-53-SiLU 100 224 75.0 9.4 27.3 M ckpt
DarkNet-Tiny 100 224 60.1 0.5 1.6 M ckpt
CSPDarkNet-Tiny 100 224 61.1 0.4 1.3 M ckpt
  • YOLOv5's Backbone
Model Epoch size acc@1 GFLOPs Params Weight
CSPDarkNet-Nano 100 224 60.6 0.3 1.3 M ckpt
CSPDarkNet-Small 100 224 69.8 1.3 4.6 M ckpt
CSPDarkNet-Medium 100 224 72.9 3.8 12.8 M ckpt
CSPDarkNet-Large 100 224 75.1 8.6 27.5 M ckpt
CSPDarkNet-Huge 100 224 16.3 50.5 M
  • YOLOv7's Backbone
Model Epoch size acc@1 GFLOPs Params Weight
ELANNet-Nano 100 224 48.7 0.03 0.4 M ckpt
ELANNet-Tiny 100 224 64.8 0.3 1.4 M ckpt
ELANNet-Large 100 224 75.1 4.1 14.4 M ckpt
ELANNet-Huge 100 224 76.2 7.5 26.4 M ckpt
  • ELANNet-v2
Model Epoch size acc@1 GFLOPs Params Weight
ELANNetv2-Pico 100 224 59.8 0.2 0.6 M ckpt
ELANNetv2-Nano 100 224 60.8 0.4 0.9 M ckpt
ELANNetv2-Tiny 100 224 67.1 0.9 1.9 M ckpt
ELANNetv2-Small 100 224 70.4 1.7 3.3 M ckpt
  • RTCNet (Yolov8's backbone)
Model Epoch size acc@1 GFLOPs Params Weight
RTCNet-P 90 224
RTCNet-N 90 224 60.7 0.38 1.36 M
RTCNet-S 90 224 1.47 4.94 M
RTCNet-M 90 224
RTCNet-L 90 224
RTCNet-X 90 224