iiinova — where your digital self keeps evolving

Sai Kumar Dwivedi @saidwivedi.in · 24/09/2026

Setup is one command. Open pockettui.com/app from any browser, enter the machine address and pairing code, and you’re in. It remembers the machine after that, so getting back to a session is one tap. [6/7]

100

Sai Kumar Dwivedi @saidwivedi.in · 24/09/2026

Everything keeps running in tmux on my machine. I can close the tab, lose the connection, or switch devices. When I come back, the same sessions are exactly where I left them. [5/7]

100

Sai Kumar Dwivedi @saidwivedi.in · 24/09/2026

I wanted this to be comfortable to use from my phone. The terminal keys I need are built in: Esc, Ctrl, Tab, arrows. For longer instructions, I can just speak. Transcription runs on my machine and uses terminal context, so it gets filenames, flags and commands right. [4/7]

100

Sai Kumar Dwivedi @saidwivedi.in · 24/09/2026

But the terminal alone doesn’t cover everything I need. Sometimes I want to browse files, check what an agent changed, read Markdown, or open an image or video a job just produced. So I built the UI I needed around the terminal, in the same browser tab. [3/7]

210

Sai Kumar Dwivedi @saidwivedi.in · 24/09/2026

PocketTUI connects to the terminal itself, so there’s nothing special to integrate. Claude Code, Codex, scripts, servers, builds, training jobs. They all stay in their real tmux sessions on your machine. If it runs in your terminal, it works with PocketTUI. [2/7]

100

Sai Kumar Dwivedi @saidwivedi.in · 24/09/2026

I use Claude Code in one terminal, Codex in another, and a training job beside them. Next month, I’ll use something else. I realized I don’t want remote access to an agent. I want remote access to where they all already live: my terminal. So I built PocketTUI.com [1/7]

130

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

InteractVLM (#CVPR2025) is a great collaboration MPI-IS, UvA and Inria. Authors: @saidwivedi.in, @anticdimi.bsky.social, S. Tripathi, O. Taheri, C. Schmid, @michael-j-black.bsky.social and @dimtzionas.bsky.social. Code & models available at: interactvlm.is.tue.mpg.de (10/10)

030

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

InteractVLM is the first method that infers 3D contacts on both humans and objects from in-the-wild images, and exploits these for 3D reconstruction via an optimization pipeline. In contrast, existing methods like PHOSA rely on handcrafted or heuristic-based contacts. (9/10)

120

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

With just 5% of DAMON’s 3D body contact annotations, InteractVLM surpasses the fully-supervised DECO baseline trained on 100% of 3D annotations. This is promising for minimizing reliance on costly 3D data by using foundational models. (8/10)

110

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

InteractVLM also shows strong outperformance on object affordance prediction on the PIAD dataset. Here affordance is defined as contact probabilities on the object. (7/10)

120

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

InteractVLM significantly outperforms prior work, both qualitatively and quantitatively, on in-the-wild 3D human (binary & semantic) contact prediction on the DAMON dataset. (6/10)

110

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

To bridge this 2D-to-3D gap, we propose "Render-Localize-Lift": - Render: 3D human/object meshes into multiview 2D images. - Localize: A Multiview Localization (MV-Loc) model, guided by VLM tokens, predicts 2D contact masks. - Lift: 2D contact masks to 3D. (5/10)

111

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

How can we infer 3D contact with limited 3D data? InteractVLM exploits foundational models—a VLM & localization model fine tuned to reason about contact. Given an image & prompt, the VLM outputs tokens for localization. But these models work in 2D, while contact is 3D. (4/10)

111

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

Furthermore, simple binary contact (touching “any” object) misses the rich semantics of real multi-object interactions. Thus, we introduce a novel task - "Semantic Human Contact" estimation: predicting contact points on a human related to a specified object. (3/10)

120

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

Precisely inferring where humans contact objects from an image is hard due to occlusion & depth ambiguity. Current datasets of images with 3D contact are small as they’re costly & tedious to create (mocap/manual labeling), limiting performance of contact predictors. (2/10)

110

Sai Kumar Dwivedi @saidwivedi.in · 15/06/2025

Why does 3D human-object reconstruction fail in the wild or get limited to a few object classes? A key missing piece is accurate 3D contact. InteractVLM (#CVPR2025) uses foundational models to infer contact on humans & objects, improving reconstruction from a single image. (1/10)

152

Sai Kumar Dwivedi @saidwivedi.in · 12/03/2025

I've been using GitHub's Lists feature for over a year, and it's seriously underrated! ⭐ It lets you assign labels to all your starred repos, making it super easy to find projects later based on specific fields or topics. No more endless scrolling! Link to my list: github.com/saidwivedi?t...

030