DenseMarks

We propose a replacement of standard face landmarks with a dense per-pixel embeddings of a full human head that live in 3D canonical space, consistent across subjects and poses. Predicted embeddings enable a variety of applications, ranging from head tracking to MVS.

Overview

DenseMarks predicts per-pixel positions in the canonical space (cube $[0, 1] ^ 3$) from human head images. The architecture is ViT + DPT head.

• Input: RGB image of size 512×512
• Output: UVW coordinate tensor (B, 3, 512, 512) with values in [0, 1]

---

🚀 Current Status

DenseMarks currently supports inference only — you can run the model to generate dense UVW predictions from input images. 🧠 Training support is coming soon! Stay tuned :)

---

Prerequisites

• Python 3.8+
• PyTorch 1.12+
• CUDA (optional, for GPU acceleration)

Installation

1. Clone the repository:

   git clone https://github.com/diddone/densemarks.git
   cd densemarks

2. Install DINOv3 submodule:

   git clone https://github.com/facebookresearch/dinov3 third_party_dinov3

3. Modify DINOv3 for compatibility:

   # For Linux (GNU sed):
   sed -i '/dinov3\.hub\.segmentors/s/^/#/; /dinov3\.hub\.classifiers/s/^/#/; /dinov3\.hub\.detectors/s/^/#/; /dinov3\.hub\.dinotxt/s/^/#/; /dinov3\.hub\.depthers/s/^/#/' third_party_dinov3/hubconf.py
   
   # For macOS (BSD sed):
   sed -i '' '/dinov3\.hub\.segmentors/s/^/#/; /dinov3\.hub\.classifiers/s/^/#/; /dinov3\.hub\.detectors/s/^/#/; /dinov3\.hub\.dinotxt/s/^/#/; /dinov3\.hub\.depthers/s/^/#/' third_party_dinov3/hubconf.py

4. Install dependencies:

   pip install torch transformers numpy

5. Download model weights from Hugging Face:

   from dense_marks_model import DenseMarksModel, read_image
   from huggingface_hub import hf_hub_download
   model = DenseMarksModel(hf_hub_download("diddone/densemarks", "model.safetensors"))
   images = read_image("assets/00000.png") # rgb, 512x512
   uvw = model(images) # Predict UVW coordinates

Citation

If you find this repo helpful, please consider cite our work:

@inproceedings{pozdeev2026densemarks,
  title={Densemarks: Learning Canonical Embeddings for Human Heads Images via Point Tracks},
  author={Dmitrii Pozdeev and Alexey Artemov and Ananta R. Bhattarai and Artem Sevastopolsky},
  booktitle={The Fourteenth International Conference on Learning Representations},
  year={2026},
  url={https://openreview.net/forum?id=KOvRxAMBzV}
}