Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/paulo-rozatto/sam-hq-exporter
Adapted SAM Exporter to SAM-HQ Exporter
https://github.com/paulo-rozatto/sam-hq-exporter
Last synced: 2 days ago
JSON representation
Adapted SAM Exporter to SAM-HQ Exporter
- Host: GitHub
- URL: https://github.com/paulo-rozatto/sam-hq-exporter
- Owner: Paulo-Rozatto
- License: mit
- Created: 2024-03-13T02:39:45.000Z (8 months ago)
- Default Branch: main
- Last Pushed: 2024-05-13T14:21:14.000Z (6 months ago)
- Last Synced: 2024-05-13T15:45:41.462Z (6 months ago)
- Language: Python
- Homepage:
- Size: 6.09 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Funding: .github/FUNDING.yml
- License: LICENSE
Awesome Lists containing this project
README
# SAM-HQ Exporter
## **Lightly adapted from [SAM Exporter](https://github.com/vietanhdev/samexporter)**
Exporting [Segment Anything in High Quality](https://github.com/SysCV/sam-hq) models to different formats.
The [Segment Anything in High Quality repository](https://github.com/SysCV/sam-hq) does not have a way to export **encoder** to ONNX format. The authors follow the original SAM's method that only export the decoder. For that reason, [@vietanhdev](https://github.com/vietanhdev) created a tool to export the original SAM model and I lightly modified it to export SAM-HQ.
**Supported models:**
- ViT-B HQ-SAM
- ViT-L HQ-SAM
- ViT-H HQ-SAM
## Installation
```bash
git clone https://github.com/vietanhdev/samexporter
cd samexporter
pip install -e .
```
## Usage
- Download all models from [Segment Anything in High Quality](https://github.com/SysCV/sam-hq?tab=readme-ov-file#model-checkpoints) repository (*.pth).
```text
original_models
+ sam_hq_vit_b.pth
+ sam_hq_vit_h.pth
+ sam_hq_vit_l.pth
...
```
- Convert encoder VIT-H HQ-SAM to ONNX format:
```bash
python -m samexporter.export_encoder --checkpoint original_models/sam_hq_vit_h.pth \
--output output_models/sam_hq_vit_h.encoder.onnx \
--model-type vit_h \
--quantize-out output_models/sam_hq_vit_h.encoder.quant.onnx \
--use-preprocess
```
- Convert decoder SAM-H to ONNX format:
```bash
python -m samexporter.export_decoder --checkpoint original_models/sam_hq_vit_h.pth \
--output output_models/sam_hq_vit_h.decoder.onnx \
--model-type vit_h \
--quantize-out output_models/sam_hq_vit_h.decoder.quant.onnx \
--return-single-mask
```
Remove `--return-single-mask` if you want to return multiple masks.
- Inference using the exported ONNX model:
```bash
python -m samexporter.inference \
--encoder_model output_models/sam_hq_vit_h.encoder.onnx \
--decoder_model output_models/sam_hq_vit_h.decoder.onnx \
--image images/truck.jpg \
--prompt images/truck_prompt.json \
--output output_images/truck.png \
--show
```
```bash
python -m samexporter.inference \
--encoder_model output_models/sam_hq_vit_h.encoder.onnx \
--decoder_model output_models/sam_hq_vit_h.decoder.onnx \
--image images/plants.png \
--prompt images/plants_prompt1.json \
--output output_images/plants_01.png \
--show
```
```bash
python -m samexporter.inference \
--encoder_model output_models/sam_hq_vit_h.encoder.onnx \
--decoder_model output_models/sam_hq_vit_h.decoder.onnx \
--image images/plants.png \
--prompt images/plants_prompt2.json \
--output output_images/plants_02.png \
--show
```
**Short options:**
In the original [SAM Exporter]() repo, there are some helper scripts that I didn't change, so I don't include them here. Their usage are as follows:
- Convert all Meta's models to ONNX format:
```bash
bash convert_all_meta_sam.sh
```
- Convert MobileSAM to ONNX format:
```bash
bash convert_mobile_sam.sh
```
## Tips
- Use "quantized" models for faster inference and smaller model size. However, the accuracy may be lower than the original models.
- SAM-B is the most lightweight model, but it has the lowest accuracy. SAM-H is the most accurate model, but it has the largest model size. SAM-M is a good trade-off between accuracy and model size.
## AnyLabeling
The original package was developed by [@vietanhdev](https://github.com/vietanhdev) for auto labeling feature in [AnyLabeling](https://github.com/vietanhdev/anylabeling) project. However, you can use it for other purposes.
[](https://youtu.be/5qVJiYNX5Kk)
I myself am using it to develop the [Subvisor](https://github.com/Paulo-Rozatto/subvisor) project. It still is a work in progress.
## License
This project is licensed under the MIT License - see the [LICENSE](LICENSE) file for details.