metadata
library_name: pytorch
SuperPoint is a self-supervised feature detection and description network that jointly learns interest point locations and corresponding descriptors for robust image matching.
Original paper: SuperPoint: Self-Supervised Interest Point Detection and Description
SuperPoint
This model uses the SuperPoint feature extractor, which detects repeatable keypoints and generates compact descriptors that are robust to viewpoint, scale, and illumination changes. It is well suited for applications such as visual localization, SLAM, structure-from-motion, image stitching, and feature-based tracking in robotics and augmented reality systems.
Model Configuration:
- Reference implementation: Official SuperPoint source code
- Original Weight: pytorch-superpoint pretrained checkpoints (Model Garden training/export flow)
- Resolution: 1x240x320 (grayscale uint8 input)
- Support Cooper version:
- Cooper SDK: [2.5.4]
- Cooper Foundry: [2.3]
| Model | Device | Compression | Model Link |
|---|---|---|---|
| SuperPoint | N1-655 | Amba_optimized | Model_Link |
| SuperPoint | N1-655 | Activation_fp16 | Model_Link |
| SuperPoint | CV7 | Amba_optimized | Model_Link |
| SuperPoint | CV7 | Activation_fp16 | Model_Link |
| SuperPoint | CV72 | Amba_optimized | Model_Link |
| SuperPoint | CV72 | Activation_fp16 | Model_Link |
| SuperPoint | CV75 | Amba_optimized | Model_Link |
| SuperPoint | CV75 | Activation_fp16 | Model_Link |