- Semantic Segmentation
Using automatic compression allows for easy model compression without a deep understanding of compression. Advanced compression offers higher flexibility in settings and may yield better compression results.
- 'Compressed' Type: Automatic Compression
- 'Compressed(Adv.)' Type: Advanced Compression
All of the original and compressed models can be downloaded easily on the Model Compressor Model Zoo.
PyTorch
| Model | Best Practice | Type | Dataset | mIoU (%) | Global Correct (%) | FLOPs (M) | Params (M) | Latency (ms) | Model Size (MB) |
|---|---|---|---|---|---|---|---|---|---|
| FCN ResNet50 | Original | COCO | 60.5 | 91.4 | 306554.91 | 35.32 | 13167.17 | 135.09 | |
| FCN ResNet50 | Google Colab | Compressed-1 | COCO | 59.6 (-0.9) | 91.4 (-0.0) | 156106.03 (1.96x) | 17.58 (2.01x) | 6438.06 (2.04x) | 67.34 (2.01x) |
| FCN ResNet50 | Google Colab | Compressed-2 | COCO | 54.7 (-5.8) | 90.7 (-0.7) | 45826.66 (x6.68) | 4.84 (7.31x) | 2147.92 (6.13x) | 18.70 (7.22x) |
- We used a subset of COCO dataset to fine-tuning FCN ResNet50. You can check more details of dataset here.
- The model's latency is measured using a Raspberry Pi 4B (1.5GHz ARM Cortex).
- Options: FP32, ONNX runtime
Updated 10 months ago