SAM stands for segmentation everything. Up to now, there are three models in this big family: SAM1, SAM2 and SAM3. I’ll give a short review about each of them. Also, I’ll introduce some variants and extensions of SAM, for example, applying stronger memory module for better tracking. Finally, I simply go through the metrics used in VOT (video object tracking) and VOS (video object segementation), then simply talk about how we apply them into our task and data. ...
During the summer semester of 2026, I took Introduction to Neural Networks and Sequence-to-Sequence Learning course. As you see, the main content is from perceptrons to modern sequence models. I started learning machine learning through Li Mu’s courses. I spent several weeks systemically studying his materials and organized the corresponding code in the repo. Also, I had already used neural networks in multiple projects before taking this course. I had trained language models, vision models, vision-language models, reinforcement-learning agents and even vision-language-action models… ...
Physical Intelligence is a fast-rising company focused on bringing general-purpose AI into the physical world. In under two years since introducing their first VLA prototype model π₀, thet’ve made a huge impact in the embodied intelligence community. In this post, I’ll walk through the three main VLA models they’ve released so far, based on my reading of their blogs and papers. π₀ π₀ is a vision-language-action (VLA) model built on top of a pre-trained vision–language model (VLM) backbone. It is then robot-pretrained on a large mixture of open-source and in-house manipulation datasets to learn broad, general skills, and can be further post-trained on smaller, task-specific data to specialize for downstream applications. ...