From the model card
Fields below are copied from the tags and counters on the HuggingFace repository depth-anything/Depth-Anything-V2-Small-hf at our last fetch. They are set by the uploader, not verified by us; rows with no tag are omitted. How this page is made.
- Publisher (HF namespace)
- depth-anything
- Pipeline tag
- depth-estimation
- Library
- Transformers
- Weight formats
- safetensors
- License tag
apache-2.0— read the license file in the repo before relying on it- Papers cited
- arXiv:2406.09414, arXiv:2401.10891
- Downloads (HF counter at last fetch)
- 1,688,778
- Likes (HF counter at last fetch)
- 46
- Model card
- https://huggingface.co/depth-anything/Depth-Anything-V2-Small-hf
Use cases
- Real-time monocular depth estimation in applications
- Depth map generation for 3D reconstruction pipelines
- Obstacle awareness for autonomous navigation
- Visual effects requiring depth-conditioned processing
Pros
- Small variant enables fast CPU and mobile GPU inference
- V2 synthetic training improves fine-grained depth accuracy over V1
- Apache-2.0 licensed
- HuggingFace-converted format for easy pipeline integration
Cons
- Monocular depth is inherently scale-ambiguous without calibration
- Small model trails Depth Anything V2 Base and Large on detailed scenes
- Not suitable as a replacement for LiDAR in safety-critical applications
- Performance degrades on transparent surfaces and mirrors
Tags
transformerssafetensorsdepth_anythingdepth-estimationdepthrelative deptharxiv:2406.09414arxiv:2401.10891license:apache-2.0endpoints_compatibleregion:us