🚨New Dataset Distillation paper 🚨 We propose PoDD, a new dataset distillation setting for a tiny🤏under 1 image-per-class (IPC) budget❗In this CIFAR100 example, current SoTA is 35.5% acc using ~100k pixels, PoDD gets 35.7% with ~40k pixels Project: vision.huji.ac.il/podd 🧵👇

Announcing our new work: lcm-lookahead.github.io 🥳 We improve IP-Adapter by training with a new identity-lookahead loss and synthetic, consistent data. The key idea is to propagate image-space losses (ID, CLIP) through LCM-LoRA instead of single-step DDPM approximations. 1/5

Can finally share that our The Chosen One paper about consistent characters generation has been accepted to #SIGGRAPH2024 🎉 More details are available here: x.com/omriavr/status… Thanks @AK for sharing 🙏

🚨New paper On the Origin of Models🚨 We define the Model Tree🌳 to describe the hereditary relations between models. We propose the task “Model Tree Heritage Recovery” (MoTHer Recovery) for discovering Model Trees via model weights Project: vision.huji.ac.il/mother 🧵👇

EffoVPR: Effective Foundation Model Utilization for Visual Place Recognition Issar Tzachor et 11 al tl;dr: 2 stage VPG with DINOv2: first retrieve with global desc, then rerank with local mutual NN matching of pre-last layer. Best when fine-tuned arxiv.org/abs/2405.18065

I am excited to share my first work: "Dataset Size Recovery from LoRA Weights". Ever wondered if you could find out how many samples was a model trained on using just its weights? Well now you can! Project: vision.huji.ac.il/dsire/ 👇

Beautiful results!

1/4 🧩 Excited to share our new paper "Visual Riddles"! We explore how small visual details can greatly impact understanding, providing a rigorous test for both visual comprehension and world knowledge factuality. 🧵

Working on AI security and safety? Consider submitting to our Neurips'24 workshop on red teaming GenAI!

Checkout our work "GameNGen". A Gaming engine powered by a diffusion-model that simulates DOOM in Real-Time! Find out more: gamengen.github.io Amazing effort and fun collaboration with the incredible Dani Valevski, Yaniv Leviathan, and Shlomi Fruchter!

Submitted an ML security paper to USENIX? Why not submit it to our workshop as well? 🪩🎉 Submission deadline September 14! Call for papers can be found on our website: redteaming-gen-ai.github.io Hope to see you in Vancouver ☃️