Keypoint Detection
Transformers
Safetensors
LightGlue
keypoint-matching
model_hub_mixin
pytorch_model_hub_mixin
Instructions to use ETH-CVG/lightglue_disk with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use ETH-CVG/lightglue_disk with Transformers:
# Load model directly from transformers import AutoImageProcessor, AutoModelForKeypointMatching processor = AutoImageProcessor.from_pretrained("ETH-CVG/lightglue_disk") model = AutoModelForKeypointMatching.from_pretrained("ETH-CVG/lightglue_disk") - Notebooks
- Google Colab
- Kaggle
Update README to inform about transformers versioning
Browse filesHey there,
we have to remove the remote code execution version in the most recent versions due to security concerns. It shouldn't have made it into the library as remote code cannot be maintained or reviewed by us in full capacity. This means that it will work with previous versions but not the following ones. Hope you understand 😄
Cheers,
Anton
README.md
CHANGED
|
@@ -10,17 +10,13 @@ pipeline_tag: keypoint-detection
|
|
| 10 |
|
| 11 |
This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:
|
| 12 |
|
| 13 |
-
This is a LightGlue variant trained on DISK, with a commecially permissive license, which requires `kornia` to be installed and is usable with transformers with the following lines of code
|
| 14 |
```python
|
| 15 |
from transformers import LightGlueForKeypointMatching
|
| 16 |
|
| 17 |
model = LightGlueForKeypointMatching.from_pretrained("ETH-CVG/lightglue_disk", trust_remote_code=True)
|
| 18 |
```
|
| 19 |
|
| 20 |
-
_Also, the commit allowing DISK to work with LightGlue is not yet included in a version of transformers, please install transformers from the main branch_
|
| 21 |
-
```
|
| 22 |
-
uv pip install git+https://github.com/huggingface/transformers.git
|
| 23 |
-
```
|
| 24 |
|
| 25 |
# LightGlue
|
| 26 |
|
|
|
|
| 10 |
|
| 11 |
This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:
|
| 12 |
|
| 13 |
+
This is a LightGlue variant trained on DISK, with a commecially permissive license, which requires `kornia` to be installed and is usable with transformers up to [v5.4.0](https://github.com/huggingface/transformers/pull/45122) with the following lines of code
|
| 14 |
```python
|
| 15 |
from transformers import LightGlueForKeypointMatching
|
| 16 |
|
| 17 |
model = LightGlueForKeypointMatching.from_pretrained("ETH-CVG/lightglue_disk", trust_remote_code=True)
|
| 18 |
```
|
| 19 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
# LightGlue
|
| 22 |
|