Наконец-то выкладываю видео моего доклада на мини-конфе в Тбилиси в марте.
Я рассказывал про основные интересности в Компьютерном Зрении (не GenAI), которые прозошли c марта 2023 до марта 2024. За пол года SOTA уже, конечно, обновилась, но концептуально я разобрал много базированных статей 23-го и 24-го года, поэтому доклад все еще актуален.
What matters in CV in 2024:
SCALE
COMPUTE
DATA
Contents:
• Visual representation learning:
• Scaling: Model & Compute & Data
• Self-supervised pre-training
• Multimodal models [briefly]
• Fine-grained tasks: Segmentation & Tracking
Papers dicsussed:
• NaViT: Vision Transformer for any Aspect Ratio and Resolution, NeurIPS 2023
• Getting ViT in Shape: Scaling Laws for Compute-Optimal Model Design, NeurIPS 2023
• ViT-22B: Scaling Vision Transformers to 22 Billion Parameters, ICML 2023
• EVA-CLIP-18B: Scaling CLIP to 18 Billion Parameters
• Data Filtering Networks (DFN), ICLR 2024
• What does clip know about a red circle? visual prompt engineering for VLMs, ICCV 2023
• SigLip: Sigmoid Loss for Language Image Pre-Training, ICCV 2023
• Image Captioners Are Scalable Vision Learners Too, NeurIPS 2023
• The effectiveness of MAE pre-pretraining for billion-scale pretraining
• DINOv2: Learning Robust Visual Features without Supervision, ICLR 2024
• ImageBind: One Embedding Space To Bind Them All, CVPR 2023
• LLaVa 1.0 & 1.5: Visual Instruction Tuning, NeurIPS 2023, Improved Baselines with Visual Instruction Tuning, arXiv 2023
• PaLI-3 Vision Language Models: Smaller, Faster, Stronger
• Segment Anything, ICCV 2023
• CoTracker: It is Better to Track Together, ECCV 2024
Ну, и на последок, вот фото со встречи эйай ньюз в Тбилиси.
https://youtu.be/Nmnl9FCXlFw
#личное #personal
@ai_newz
Please open Telegram to view this post
VIEW IN TELEGRAM
YouTube
Computer Vision Research in 2023-2024: A Brief Overview
Recording of the talk by Artsiom Sanakoyeu from Opentalks, Tbilisi, March 2024.
X: twitter: x.com/artsiom_s
Talk abstract:
In this talk I will spotlight the year's (2023-March 2024) most exciting papers and advancements in Computer Vision. From novel scaled…
X: twitter: x.com/artsiom_s
Talk abstract:
In this talk I will spotlight the year's (2023-March 2024) most exciting papers and advancements in Computer Vision. From novel scaled…
2👍77❤38🔥16😁2😱2🦄2🤩1