Yuecheng919

Yuecheng919 / GemDepth

Public

【ICML 2026】GemDepth: Geometry-Embedded Features for 3D-Consistent Video Depth

43
2
100% credibility
Found May 15, 2026 at 59 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

GemDepth estimates temporally consistent depth maps and point clouds from monocular videos using geometry-aware features.

How It Works

1
🔍 Discover GemDepth

You find a cool tool that turns everyday videos into 3D depth maps with smooth motion.

2
📥 Get it ready

Download and set up the tool on your computer in a few simple steps.

3
🎥 Pick your video

Choose a video from your phone or camera to explore its hidden 3D world.

4
✨ Create 3D depth

Hit go and watch as the tool reveals detailed depth layers that stay consistent across frames.

5
📹 View depth video

Play back your video with colorful depth overlays showing near and far.

6
Make point clouds?
✅
Yes, create clouds

Generate colorful 3D points you can rotate and explore.

❌
Just depth video

Stick with the smooth depth video results.

🎉 3D magic unlocked

Your video now has lifelike depth and motion, ready for AR, editing, or fun 3D views!

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 59 to 43 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is GemDepth?

GemDepth is a Python tool for estimating depth from monocular videos with strong 3D consistency. It injects camera motion and geometry priors to deliver sharp per-frame depths that stay stable across time, fixing flicker in dynamic scenes. Run inference scripts on video folders to get depth maps, visualizations, or pointclouds—plus eval on KITTI, Sintel, Bonn, ScanNet.

Why is it gaining traction?

SOTA benchmarks show it crushes alternatives in temporal consistency and detail recovery, especially motion-heavy clips. Dead-simple CLI like `run_video.py` or `run_video_pointcloud.py` spits out results fast, no fuss. As an ICML 2026 github project with arXiv paper, it's drawing eyes from icml papers github crowd chasing geometry-embedded video depth.

Who should use this?

Computer vision researchers benchmarking 3d-consistent depth on icml 2025 github datasets or dynamic videos. Robotics devs needing stable monocular depth for SLAM or navigation. AR/VR builders prototyping pointcloud reconstruction from phone footage.

Verdict

Solid pick for video depth with geometry-embedded features, but 43 stars and 1.0% credibility score scream early days—light docs, no tests. Download the HF model, test inference; track for ICML 2026 polish.

(178 words)

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.