
CS Ph.D. student @UMD. Interested in Video Understanding.
College Park, MD
Joined March 2021
- Tweets50
- Following99
- Followers80
- Likes249
🧵 VidParse has been accepted to #ECCV2026! 🎉
Can we parse a how-to video as it unfolds - without training?
Yes. And it beats the best online supervised baselines, while coming close to offline performance. 🚀
Paper 👇
arxiv.org/abs/2608.27562
👥 This work was done with @architkam, @vatsalag_2, @pulkitkumar95 & @abhi2610
We're releasing everything:
🌐 Project: learn2phoenix.github.io/VidP…
📄 Paper: arxiv.org/abs/2608.27562
💻 Code: github.com/learn2phoenix/Vid…
Anubhav Gupta retweeted
We introduce LEIA, a novel approach for representing 3D articulated objects by learning view-invariant latent embeddings for different articulation states. Visit our poster #299 at @eccvconf #ECCV2024 in Milan on Oct 3rd (Thursday), 10:30am.
Project Page: archana1998.github.io/leia/
Anubhav Gupta retweeted
🖼️How to measure whether a style exists in a generated image? If you want to know more about it, drop by our poster tomorrow at #ECCV2024 ! We also released model checkpoints and the training data! ✨
⏰: Oct 1 (Tuesday) 10.30 session
📍- Poster # 111
💻 - somepago.github.io/csd.html
#diffusion #GenAI
✨ Can we detect style in the generated images? Our recent work takes a step towards understanding this question. We train a style-focused vision feature extractor built on top of CLIP, which we call a Contrastive Style Descriptior (CSD).
paper: arxiv.org/abs/2404.01292
Style similarity on CSD vectors > 0.8 more or less indicates the existence of style. And similarity less than < 0.5 indicates no style. In this figure, we can see, some artists' styles (Greg Rutkwoski, Amano) are clearly removed in the training of SD 2.1, while they are still present in SD 1.4 and Midjourney.
On the other hand, we can see a strong presence of style of artists such as Afremov, Seurat, Mucha in all 3 models.
#diffusion #MachineLearning
1/8 🧶
Anubhav Gupta retweeted
Measuring Style Similarity in Diffusion Models
Generative models are now widely used by graphic designers and artists. Prior works have shown that these models remember and often replicate content from their training data during generation. Hence as their proliferation
Anubhav Gupta retweeted
📢Introducing ClimateLearn, a new PyTorch library for accessing climate datasets, state-of-the-art ML models, and high quality training and visualization pipelines.
Blog: aditya-grover-group.github.i…
Docs: climatelearn.readthedocs.io/
Quickstart Colab: colab.research.google.com/dr…
🧵 (1/n)
Anubhav Gupta retweeted
Survey on Large Scale Neural Network Training
A nice summary of strategies and insights on large scale neural network training. Great read for ML engineers interested in training efficient DNNs.
arxiv.org/abs/2202.10435
Anubhav Gupta retweeted
Annotated (with notes) implementation of Rotary Positional Embeddings (RoPE)
📝 Code/notes nn.labml.ai/transformers/rop…
It introduces a new form of positional encodings for transformers that encodes position in a relative form. Models seem to converge faster with RoPE. (1/4)
🧵👇
Anubhav Gupta retweeted
Transformer Memory as a Differentiable Search Index
Shows that that information retrieval can be done with a single Transformer, in which all information about the corpus is encoded in the parameters of the model.
arxiv.org/abs/2202.06991
Anubhav Gupta retweeted
We are releasing MOTSynth, the largest dataset for Pedestrian Detection and Tracking.
Expect to obtain better real-world performances by training your model on this data only.
It's synthetic, but it's for real! lnkd.in/d-NvqiX4
Anubhav Gupta retweeted
🔥The WILDS 2.0 benchmark is an oral at ICLR'22! Existing distribution shift benchmarks do not reflect the breadth of scenarios that arise in real-world applications. Wilds 2.0 extends the Wilds benchmark to include curated unlabeled data for DA research.
arxiv.org/abs/2112.05090
Anubhav Gupta retweeted
We computer vision researchers rarely look at the individual data points inside our datasets.
Mainly because we are too lazy and/or do not have the right tools.
This needs to change. And now we have a great tool from @TensorFlow datasets team: Know Your Data. A thread.🧵(1/5)
Anubhav Gupta retweeted
A new study from @UMD_AGNR has found that some sectors, like agriculture, have barely scratched the surface to identify and address potential ethical, legal, social, and economic implications of artificial intelligence.
Read the full story: go.umd.edu/qMo





















