[Transformer] by Hand ✍️
Make Your Own 🛠️👉 https://t.co/ZuXhITccaL
Over the past few months, I've collaborated with several AI educators to customize my AI by Hand exercises. I am glad that my materials are being used and appreciated in many classrooms around the world!
However, because the customization process is entirely by hand, sometimes my solution contains errors, only to be caught by students, which actually made me happy that students are paying attention. 😅
Lately, I have been thinking about developing a tool to let people create their own AI by Hand exercises, with custom numbers and solutions.
After considering a range of technologies, I decided to use Google Sheets. My goal is to maximize reach and broaden access.
Because this tool is still in the early stage, I would really appreciate your feedback!
What other topics do you want to see next?
视觉-语言模型(VLM)领域在研究些什么?🧐
VLM是一个从去年末开始快速发展的领域,对研究者来说尚有大量“金矿”未被发掘,且当前探索仍然非常初步,对大模型的初学者上手难度较小🥰
以下是帮你快速掌握VLM领域目前发展的文章推荐📰:
1. 从宏观视角整体了解整个领域有哪些具体的探索方向(例如数据配比、Image Encoder选择、VL connector的设计、当前有哪些benchmark、VLM的训练策略等)
a. Cambrian: A Fully Open, Vision-Centric Exploration of Multimodal LLMs
最全最新没有之一的全方位探索
Link: https://t.co/fqS9zVB5AS
b. MM1: Methods, Analysis & Insights from Multimodal LLM Pre-training
比较老但仍然推荐一读的文章
Link: https://t.co/Q7HSytHhwB
c. What matters when building vision-language models?
结论相比前两篇有很好的补充
Link: https://t.co/xOTVQj8PZ6
2. VLM特有的提升推理效率方案:设计更优的V-L Attention机制
a. An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models
发现vision token存在大量的冗余,可以通过token dropping来大幅提升推理速度而不伤害效果
Link: https://t.co/4hvsgy0nr7
b. VoCo-LLaMA: Towards Vision Compression with Large Language Models
通过类似RMT的token压缩方式减少vision token数量从而提升推理速度
Link: https://t.co/8227vL5Sd0
3. vision encoder的分辨率对模型性能的影响,结论简单粗暴:影响很大,分辨率越大效果越好
a. InternLM-XComposer2-4KHD: A Pioneering Large Vision-Language Model Handling Resolutions from 336 Pixels to 4K HD
Link: https://t.co/hcUik3aVlQ
b. DeepSeek-VL: Towards Real-World Vision-Language Understanding
Link: https://t.co/OBJRnRBlK1
4. VLM模型架构选择:All-in-one Decoder (early-fusion)还是Vision Encoder和Language Decoder分离?
a. Unveiling Encoder-Free Vision-Language Models
https://t.co/puhPIluZEB
b. Chameleon: Mixed-Modal Early-Fusion Foundation Models
https://t.co/p5iWwWHbMA
5. 对于较为主流的VLM分离架构,Vision-Language Connector如何设计?
a. TokenPacker: Efficient Visual Projector for Multimodal LLM
https://t.co/s3N1ntrajw
6. VLM分离架构的最佳训练方式
a. Long Context Transfer from Language to Vision
https://t.co/eoKGiw9ufm
7. LLaVA系列的所有文章+博客
Improved Baselines with Visual Instruction Tuning
https://t.co/8pLH4RNxAZ
https://t.co/amvIg2TynU
https://t.co/cfVN0Pf6at
https://t.co/CKiCyN2d0G
https://t.co/By6hrZyNyU
https://t.co/PMX6iqmwHt
8. 一些快速提升你VLM码力的实战仓库推荐(见图)
(列得不够全希望大家在评论区继续补充)
An important update on the #CVPR2024 submission deadline from the conference organizing committee:
Our Program Chairs have voted to shift the CVPR 2024 submission deadline to November 17th (A one-week extension). The website will be updated shortly to reflect this change.
The number of papers does not matter.
- professors who always tweet "N (N>10) papers got into CVPR/ICCV/SIGGRAPH/NeurIPS..." and received tens of "congrats" afterwards
We are pleased to share our work on monocular depth estimation and single-image 3D reconstruction, Metric3D, which was accepted by ICCV2023. This method won first place in the 2nd Monocular Depth Estimation Competition at CVPR.
Arxiv:https://t.co/QoOSKrW2Bo