metadata
library_name: pytorch
FSRCNN is a lightweight super-resolution network that accelerates single-image upscaling by performing feature extraction and reconstruction directly in the low-resolution space, significantly reducing computational cost compared to earlier approaches.
Original paper: Accelerating the Super-Resolution Convolutional Neural Network
FSRCNNx4
This model uses the FSRCNN ×4 variant, which is trained to reconstruct a high-resolution image at four times the input resolution. It is well suited for applications such as image enhancement, video upscaling, surveillance imagery, and edge devices where fast super-resolution inference is required.
Model Configuration:
- Reference implementation: FSRCNN
- Original Weight: FSRCNNx4_Weights.91-image
- Resolution: 1x1x128x128
- Support Cooper version:
- Cooper SDK: [2.5.4]
- Cooper Foundry: [2.3]
| Model | Device | compression | Model Link |
|---|---|---|---|
| FSRCNNx4 | N1-655 | Amba_optimized | Model_Link |
| FSRCNNx4 | N1-655 | Activation_fp16 | Model_Link |
| FSRCNNx4 | CV7 | Amba_optimized | Model_Link |
| FSRCNNx4 | CV7 | Activation_fp16 | Model_Link |
| FSRCNNx4 | CV72 | Amba_optimized | Model_Link |
| FSRCNNx4 | CV72 | Activation_fp16 | Model_Link |
| FSRCNNx4 | CV75 | Amba_optimized | Model_Link |
| FSRCNNx4 | CV75 | Activation_fp16 | Model_Link |