| S. No | File Name | Description |
|---|---|---|
| 1 | anyloc_vlad_generate_colab.ipynb | Jupyter notebook (for Google Colab) to generate global descriptors using AnyLoc-VALD-DINOv2. |
| 2 | anyloc_vlad_generate.py | Python script to generate global descriptors using AnyLoc-VALD-DINOv2. |
| 3 | images_vlad_colab.ipynb | WIP: Jupyter notebook (for Google Colab). Visualize cluster assignments. |
| 4 | hf_imgs_vlad_clusters.py | WIP: HuggingFace app for visualizing the cluster assignments. |
You'll need the following before getting started. The Colab notebooks actually setup all this (as a part of the notebook), they're fully standalone.
- Cluster Centers: Download the cluster centers from the public data:
Colab1 > cache.zip. Unzip the file inpwdwithunzip cache.zipso that./cacheis created here. Vocabulary is stored inc_center.ptfiles in folders like./cache/vocabulary/dinov2_vitg14/l31_value_c32/urban.
If you want to use the python script (after cloning the repo and setting up utilities).
python ./anyloc_vlad_generate.py
# Or if you have your own *.png images
python ./anyloc_vlad_generate.py --in-dir ./data/images --imgs-ext png --out-dir ./data/descriptorsUse --domain to specify the domain of the images (default is urban). Use --help to see all the options.
Tip: You can get an idea of which domain to use by using our HuggingFace Space app. Upload a representative sample of your images under
GeM t-SNE Projectionand see which group of images do your uploaded images come close to.
