A free, open-source library for training PyTorch-based computer vision models. Developed by Deci's deep learning experts for the benefit of the AI community.
A training script for all the most commonly applied computer vision tasks
Recipes for quick and simple reproduction of SOTA model results
Plug & Play Hyperparameters for training, dataset and architecture
Easily load and fine-tune production-ready, pre-trained SOTA models that incorporate best practices and validated hyper-parameters for achieving best-in-class accuracy.
E.g YOLOv5, DDRNet, EfficientNet, RegNet, ResNet, MobileNet, etc.
Why do all the grind work, if we already did it for you? leverage tested and proven recipes & code examples for a wide range of computer vision models generated by our team of deep learning experts. Easily configure your own or use plug & play hyperparameters for training, dataset, and architecture. The models were also validated for compatibility with various graph compilers.
Need to train models on different tasks? Now, you no longer need to scout through different repositories to find the most suitable architecture for your needs - the SuperGradients repo has them all in one place.
Classification | Detection | Segmentation
Forget about dealing with different and stiff formats and conventions while trying to integrate training libraries into your code base. With SuperGradients integration is simple and straightforward.
| Model | Dataset | Resolution | Top-1 | Top-5 | Latency Batch size 1 T4 GPU | Throughput Batch size 1 T4 GPU |
|---|---|---|---|---|---|---|
| EfficientNet B0 | ImageNet | 224×224 | 77.62 | 93.49 | 1.16ms | 862fps |
| RegNetY200 | ImageNet | 224×224 | 70.88 | 89.35 | 1.07ms | 928.3fps |
| RegNetY400 | ImageNet | 224×224 | 74.74 | 91.46 | 1.22ms | 816.5fps |
| RegNetY600 | ImageNet | 224×224 | 76.18 | 92.34 | 1.19ms | 838.5fps |
| RegNetY800 | ImageNet | 224×224 | 77.07 | 93.26 | 1.18ms | 841.4fps |
| ResNet18 | ImageNet | 224×224 | 70.6 | 89.64 | 0.599ms | 1669fps |
| ResNet34 | ImageNet | 224×224 | 74.13 | 91.7 | 0.89ms | 1123fps |
| ResNet50 | ImageNet | 224×224 | 76.3 | 93 | 0.94ms | 1063fps |
| MobileNetV3_large-150 epochs | ImageNet | 224×224 | 73.79 | 91.54 | 0.87ms | 1149fps |
| MobileNetV3_large-300 epochs | ImageNet | 224×224 | 74.52 | 91.92 | 0.87ms | 1149fps |
| MobileNetV3_small | ImageNet | 224×224 | 67.45 | 87.47 | 0.75ms | 1333fps |
| MobileNetV2_w1 | ImageNet | 224×224 | 73.08 | 91.1 | 0.58ms | 1724fps |
| Model | Dataset | Resolution | mAPval 0.5:0.95 | Latency Batch size 1 T4 GPU | Throughput Batch size 64 T4 GPU |
|---|---|---|---|---|---|
| YOLOv5 nano | COCO | 640×640 | 27.7 | 6.55ms | 177.62fps |
| YOLOv5 small | COCO | 640×640 | 37.3 | 7.13ms | 159.44fps |
| YOLOv5 medium | COCO | 640×640 | 45.2 | 8.95ms | 121.78fps |
| YOLOv5 large | COCO | 640×640 | 48 | 11.49ms | 95.99fps |
| Model | Dataset | Resolution | mIoU | Latency Batch size 1 T4 GPU | Throughput Batch size 1 T4 GPU | Latency Batch size 1 T4 GPU including IO |
|---|---|---|---|---|---|---|
| DDRNet23 | Cityscapes | 1024×2048 | 78.65 | 7.62ms | 131.3fps | 25.94ms |
| DDRNet23 slim | Cityscapes | 1024×2048 | 76.6 | 3.56ms | 280.5fps | 22.80ms |
| STDC1-Seg50 | Cityscapes | 512×1024 | 74.36 | 2.83ms | 353.3fps | 12.57ms |
| STDC1-Seg75 | Cityscapes | 768×1536 | 76.87 | 5.71ms | 175.1fps | 26.70ms |
| STDC2-Seg50 | Cityscapes | 512×1024 | 75.27 | 3.74ms | 267.2fps | 13.89ms |
| STDC2-Seg75 | Cityscapes | 768×1536 | 78.93 | 7.35ms | 135.9fps | 28.18ms |
| ShelfNet_LW_34 | COCO Segmentation | 512×512 | 65.1 | – | – | – |
Stay informed and get the latest product updates, community announcements, and events.
This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.
The Deep Learning Daily Discord community is a dedicated space for meeting deep learning experts. Here you can ask and answer questions, have open-ended conversations and stay up to date on everything deep learning.
