Efficient Track Anything is accepted at #ICCV2025!!.

Drop by our oral presentation and poster session to chat and learn about our video tokenizer with learned autoregressive prior. #ICLR2025.

RT @AIatMeta: Today is the start of a new era of natively multimodal AI innovation. Today, we’re introducing the first Llama 4 models: Lla….

📢 Excited to announce LARP has been accepted to #ICLR2025 ! 🇸🇬 .Code and models are publicly available. Project page:

RT @fiandola: [1/n] 𝗘𝗳𝗳𝗶𝗰𝗶𝗲𝗻𝘁 𝗧𝗿𝗮𝗰𝗸 𝗔𝗻𝘆𝘁𝗵𝗶𝗻𝗴 from @Meta: interactive video segmentation and tracking on an iPhone!

Checkout Efficient Track Anything from our team. 2x faster than SAM2 on A100 .> 10 FPS on iPhone 15 Pro Max. Paper: demo:

Checkout LARP, our work on creating a video tokenizer which is trained with an autoregressive generative prior. Code and models are open sourced!.

We are happy to release our LiFT code and pretrained models! 📢. Code: Project Page: Here are some super spooky super resolved feature visualizations to make the season scarier 🎃. Coauthors: @MatthewWalmer @kamalgupta09 @abhi2610

RT @YoungXiong1: 🚨VideoLLM from Meta!🚨.LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding. 📝Paper: https://t….

Work done with @MatthewWalmer, @kamalgupta09 and @abhi2610 .

We see consistent gains across multiple tasks which require dense features. On applying it 4x it can generate crisp image resolution features.

LiFT is trained to generate 2x higher resolution features. It also uses the low resolution image to guide the upsampling. Further due to its modular design, it can be reapplied to its own outputs.

LiFT is a small module consisting of convolution and deconvolution layers and is trained with a self-supervised reconstruction loss over the features.

We introduce LiFT, an easy to train, lightweight, and efficient feature upsampler to get dense ViT features without the need to retrain the ViT. Visit our poster @eccvconf #eccv2024 in Milan on Oct 1st (Tuesday), 16:30 (local), Poster: 79. Project Page:

Excited to announce that I have joined @AIatMeta as a Research Scientist where I will be working on model optimization. Also I will be at ECCV to present my work and am excited to meet and learn from everyone. Reach out if you are attending and would like to chat. Ciao 🇮🇹.

That's a wrap! Happy to share that I have defended my thesis. Thankful for the insightful questions and feedback from my committee members @abhi2610,@zhoutianyi, @davwiljac, Prof. Espy-Wilson, and Prof. Andrew Zisserman.

RT @deedydas: Thank you King Kohli 👑 for all the memories. End of an era.

RT @abhi2610: Call for Papers: #INRV2024 Workshop on Implicit Neural Representation for Vision @ #CVPR2024! Topics: Compression, Representa….

RT @AnthropicAI: Today, we're announcing Claude 3, our next generation of AI models. The three state-of-the-art models—Claude 3 Opus, Cla….

RT @StabilityAI: Announcing Stable Diffusion 3, our most capable text-to-image model, utilizing a diffusion transformer architecture for gr….