From 8 -> 5 -> 4 exocentric cameras, all visualized with @rerundotio. I'm dropping the number of cameras used and collecting my own data to make sure I'm not overfitting to open-source datasets.

It's finally done, I've finished ripping out my full-body pipeline and replaced it with a hands-only version. Critical to make it work in a lot more scenarios! I've visualized the final predictions with @rerundotio!

If you're not labeling your own data, you're NGMI. I take this seriously, so I finished building the first version of my hand-tracking annotation app using rerun.io and gradio.app

Unfortunately, I've identified some serious issues with the hand tracker that necessitate manual intervention. So I decided the next best course of action is to build a labeling app with @rerun.io and @gradio-hf.bsky.social, as I transition from my full-body solution to a hands-only solution.

One of the failure modes for my pipeline is the full-body tracker. This is a huge issue for ego views, where it falls apart. I've been looking for a solution that generalizes to ego/exo, and I think I found it!

Go check out ethz-vlg. github. io/mvtracker/it's a really clever idea for multiview tracking, and they made an awesome @rerun.io‬ integration on their project page!

This is a big one! I'm almost done with my MVP, built with @rerun.io‬ and @gradio-hf.bsky.social‬! TLDR: Inputs - 2-N Synchronized Videos in a zip file Outputs - Multiview Consistent Depthmaps + Pointclouds (Thanks VGGT + MogeV2!) - Depth Confidence Values

✨ Massive Pipeline Refactor → One Framework for Ego + Exo Datasets, Visualized with @rerun.io 🚀 After a refactoring, my entire egocentric/exocentric pipeline is now modular. One codebase handles different sensor layouts and outputs a unified, multimodal timeseries file that you can open in Rerun.

Streaming iPhone data in real-time directly to @rerun.io 🚀

MVP of Multiview Video → Camera parameters + 3D keypoints. Visualized with @rerun.io

First signs of life on my self-collected data =] I still need to add synchronization between the exo/ego videos, but all of the work the past few months is coming together This shows a synchronization app I'm building with @rerun.io and @gradio-hf.bsky.social . More to come

Trying to wrap my head around fwd/bwd kinematics for imitation learning, so I built a fully‑differentiable kinematic hand skeleton in JAX and visualized it with @rerun.io new callback system in a Jupyter Notebook. This shows each joint angle and how it impacts the kinematic skeleton.

I added export settings to get the images/masks/camera parameters in NerfStudio format. I can chuck these into brush from @arthurperpixel.bsky.social and get a splat on my M4. Right now, such sparse input doesn't produce great results, but the depths are not being used, which should be able to help

@pablovelagomez.bsky.social · last yr.

@rerun.io v0.23 is finally out! 🎉 I’ve extended my @gradio-hf.bsky.social annotation pipeline to support multiview videos using the callback system introduced in 0.23.

@rerun.io v0.23 is finally out! 🎉 I’ve extended my @gradio-hf.bsky.social annotation pipeline to support multiview videos using the callback system introduced in 0.23.

Visualized with @rerun.io, I’ve integrated video‑based depth estimation into my robot‑training pipeline to make data collection as accessible as possible—without requiring specialized hardware.

I extended my previous @rerun.io and @gradio-hf.bsky.social annotation pipeline for multiple views. You can see how powerful this is when using Meta's Segment Anything and multi-view geometry. Only annotating 2 views, I can triangulate the other 6 views and get masks extremely quickly!

@pablovelagomez.bsky.social · last yr.

Here’s a sneak peek using @rerun.io and @gradio-hf.bsky.social for data annotation. It uses Video Depth Anything and Segment Anything 2 under the hood to generate segmentation masks and depth maps/point clouds. More to share next week.

Here’s a sneak peek using @rerun.io and @gradio-hf.bsky.social for data annotation. It uses Video Depth Anything and Segment Anything 2 under the hood to generate segmentation masks and depth maps/point clouds. More to share next week.

Continuing with my robot training collection pipeline, one of the challenges has been obtaining calibrated cameras from sparse multi-view inputs. I previously worked with Dust3r but found its accuracy insufficient for generating reliable camera parameters.

@pablovelagomez.bsky.social · 2y ago

More progress towards building a straightforward method to collect first-person (ego) and third-person (exo) data for robotic training in @rerun.io. I’ve been using the HO-cap dataset to establish a baseline, and here are some updates I’ve made (code at the end)

More progress towards building a straightforward method to collect first-person (ego) and third-person (exo) data for robotic training in @rerun.io. I’ve been using the HO-cap dataset to establish a baseline, and here are some updates I’ve made (code at the end)

Finally finished porting mast3r-slam to @rerun.io and adding a @gradio-hf.bsky.social interface. Really cool to see it running on any video I throw at it, I've included the code at the end

I'm working towards an easy method to collect a combined third-person and first-person pose dataset starting Assembly101 from Meta, with near real-time performance via @rerun.io visualization. The end goal is robot imitation learning with Hugging Face LeRobot

Following up on my prompt depth anything post, I'm starting a bit of a miniseries where I'm going through the tutorials of Lerobot to understand better how I can get a real robot to work on my custom dataset. Using @rerun.io to visualize code: github.com/rerun-io/pi0...

Recently, I've been playing with my iPhone ToF sensor, but the problem has always been the abysmal resolution (256x192). The team behind DepthAnything released PromptDepthAnything that fixes this. Using @rerun.io to visualize. Links at the end of the thread