iiinova — where your digital self keeps evolving

Ishan Khatri @ishan.khatri.io · 23/09/2026

And OV3D-Bench is a re-evaluation of 3D monocular object detectors. Honestly a reckoning that’s a long time coming. I love this field but this “target aware” evaluation crap needed to go. You can’t prompt with only classes that are present in the image!!! This hides the false positives…

100

Ishan Khatri @ishan.khatri.io · 23/09/2026

Quick belated paper promotion. We had 2 papers at ECCV that I was in Malmo to present! UniFlow is a foundation model for 3D LiDAR scene flow, demonstrating zero-shot generalization across datasets and LiDAR sensors for the first time. We also showed that semantics don’t generalize the same way.

141

Ishan Khatri @ishan.khatri.io · 29/07/2026

Yeah this is a banger format

1513

Ishan Khatri @ishan.khatri.io · 14/06/2026

It’s a beautiful day to make some GPUs go brr

Picture of a MacBook outside, the terminal is open on the screen and 4x GB200 are visible.

060

Ishan Khatri @ishan.khatri.io · 10/04/2026

The exact number really doesn't matter here, but it's super nice to see that people are actually reading the things I have worked on :)

050

Ishan Khatri @ishan.khatri.io · 04/02/2026

@eugenevinitsky.bsky.social this reminds me of a legendary post made by my TA (and future collaborator) Kyle. Clearly you’re a lot nicer than my instructors ever were 😂

A piazza post where the students are asking for an autograder with unit tests and the TAs have responded by savagely roasting the students for asking, emphasizing that in the real world you must write your own unit tests.

140

Ishan Khatri @ishan.khatri.io · 17/01/2026

Too late! :P I used to daily arch in college it’ll be fine.

100

Ishan Khatri @ishan.khatri.io · 05/08/2025

This is a real rebuttal from the papers I am reviewing at NeurIPS... the reviewer who wrote this gave the paper a 2 (reject) with a confidence of 5 (absolutely certain & familiar with all related work, the highest confidence available). What the hell are we even doing here...?

An image showing a reviewer comment of "paper does not address object detection" with the authors responding with a list of object detection metrics included in the paper...

190

Ishan Khatri @ishan.khatri.io · 05/07/2025

JL Lemma mentioned! 😍

A paragraph from a paper mentioning the Johnson-Lindenstrauss lemma

000

Ishan Khatri @ishan.khatri.io · 11/06/2025

On my way to @cvprconference.bsky.social! Catch our presentation on the 2025 Argoverse challenges at the Workshop on Autonomous Driving tomorrow at 11:30am in Grand C1!

000

Ishan Khatri @ishan.khatri.io · 22/04/2025

Given a natural language text prompt Ex: "vehicles slowing down for a jaywalker" the goal is to identify, detect & track the relevant actors in the scene. All sensor & map data can be used, and we encourage teams to use pre-trained models as well.

110

Ishan Khatri @ishan.khatri.io · 22/04/2025

Just over a month left to submit to this year's Argoverse 2 challenges! Returning from previous years, are our motion forecasting and lidar scene flow challenges. And NEW for this year with a $10k prize pool is our Scenario Mining challenge!🧵👇

141

Ishan Khatri @ishan.khatri.io · 19/04/2025

On my way to Singapore for ICLR! If you want to chat about vision or robots, hit me up!

030

Ishan Khatri @ishan.khatri.io · 06/04/2025

Was cleaning out some random papers from my desk and found this shipping receipt from @bhphoto.com from last year when they sent me my SD card that I left in a camera I returned. I didn’t even realize I left it in there until it showed up at my doorstep. This is why I’ll be a B&H customer for life.

140

Ishan Khatri @ishan.khatri.io · 27/01/2025

MOE really refers to making a certain part of the architecture sparse to save computation. I think that “sparsity” is a better layman’s explanation than “topic experts”. There’s a really good example in this blog. newsletter.maartengrootendorst.com/p/a-visual-g...

A table from the paper ST-MoE showing which kinds of experts activated on which kinds of input tokens.

An image showing tokens colorized by which “expert” FFN was activated when it was interpreted by a transformer based LLM.

3303

Ishan Khatri @ishan.khatri.io · 13/01/2025

Ahh nvm I am stupid. There is an option for "official comment" which lets you leave a comment for ACs and PCs only. Screenshot attached if anyone else is having trouble finding it. This page exists per paper so you have to click "submit review" from the reviewer console first.

OpenReview page showing how to submit an official comment to PCs and ACs

010

Ishan Khatri @ishan.khatri.io · 28/12/2024

Lots of anti-Indian, anti-immigrant sentiment in the zeitgeist recently. Will use this opportunity to post my 🔥 outfit from a wedding this past week. Plenty of my non-Indian friends wearing traditional Indian clothing as well and absolutely killing it. Why be racist when you could look this good?

Me wearing a traditional Indian Kurta & Nehru jacket with oxfords.

4310

Ishan Khatri @ishan.khatri.io · 19/12/2024

Unfortunately this is the default ML experience.

Image of a tweet where I complain about dependency management in ML and joke that I’ll get SOTA by gluing things together because I’m the only one willing to put up with resolving the dependency conflicts.

2262

Ishan Khatri @ishan.khatri.io · 29/11/2024

Sure, the company could set some TOS (though see the image for how well that works lmao) but the whole point of AT proto is that it's a protocol. Anyone can make a service that contributes to (and reads from) the firehose.

100

Ishan Khatri @ishan.khatri.io · 23/11/2024

Also not to toot my own horn too much but we did this in ZeroFlow and we have a whole section about the significance of errors compared with the training variance over 3 training runs. It does cost compute but more people should do this!

Section of a paper describing the training variance of a method

051

Ishan Khatri @ishan.khatri.io · 21/11/2024

In the left hand sidebar there’s a section called lists which does this (I haven’t tried it yet):

A screenshot of the “lists” section of the bluesky app

100

Ishan Khatri @ishan.khatri.io · 11/11/2024

This image is on Chris Amato's Dec-POMDP page which has some excellent explanations. rbr.cs.umass.edu/camato/decpo...

020

Ishan Khatri @ishan.khatri.io · 09/11/2024

Researchers please, I am begging. Pleading. Run your code through mypy or another static analysis tool just once. These kinds of mistakes are so trivially caught. I know that grad student time isn’t worth much, but mypy, pyright, and beartype (to name a few) are all free 🥹

010

Ishan Khatri @ishan.khatri.io · 17/09/2024

Lmao yeah it was a Wild West out there in the beginning.

010

Ishan Khatri @ishan.khatri.io · 05/05/2023

I believe someone has already done this lol.

140

Ishan Khatri @ishan.khatri.io · 05/05/2023

Yeah, sorry if I explained it poorly. Here’s a screenshot from joinmastodon.org

120