I was working on semantic segmentation. used to spend a long time creating graphics for presenting at conferences. I had a link showing the results but people were saying i was sharing too many links so deleted the link. But these tools with chatgpt can write a paper in a week which i used to take 6 months to do
I have been using different anonimization tools to blur out my face and license plate at as many locations as possible to keep my data as private as possible. This recent tool does it in a single shot in a chat. I was able to blur out LPR data very easily with high accuracy https://chat.vlm.run/c/7ca96025-1d6c-4c33-ab82-97e6f017883b
yeah the video models still do not understand physics the way humans do. We are getting there one step at a time. By the way, I am seeing a lot of people complain about google billing not working well. I was able to generate these for free without signing. Look at the results and try to come up with your own failure and working use cases.
Claude, gemini and ChatGPT does image segmentation in surprising ways - we did a small evaluation [1] of different frontier models for image segementation and understanding, and Claude is by far the most surprising in results.
I spend years working on training these models. Inference is always the fruit. The effort going into getting the data is the most time consuming part. I am not a fan of meta from a long time. But open sourcing the weights help move the field in general. So I have to be thankful for that.
If Nano-Banana-pro with Veo 3.1 existed during my PhD, I would’ve finished a 6-year dissertation in a single year — it’s generating ideas today that used to take me 18 months just to convince people were possible.
yeah sadly veo 3.1 has not caught up to the image generation capabilities. May be we need to work on how to make video generation more physically consistent. but the image generation results from banana pro are great.
yes they are pricey but the price will go down over time and then you can switch. vlm.run got access as early customers and are releasing it for free with unlimited generations(till they are bottlenecked by google). some results here combining image gen(Nano Banana pro) with video gen(Veo 3.1) in a single chat https://chat.vlm.run/c/1c726fab-04ef-47cc-923d-cb3b005d6262. This combined the synth generation of a person and made the puppet dance. Quite impressive
The problem is the data. LLM data is self supervised. Vision data is very sparsly annotated in the real world. Going a step further robotics data is is much sparser. So getting these models to improve on this long tail distribution will take time.
Not of fan of the company for the social media but have to appreciate all the open sourcing. none of the other top labs release thier models like meta.
Agreed that Unet has been the most used model for medical imaging for the last 10 years since the initial Unet paper. I think a combination of Llm+VLMs could be a way forward for medical imaging. I tried it out here and it works great. https://chat.vlm.run/c/e062aa6d-41bb-4fc2-b3e4-7e70b45562cf
I just check and it seems to commercial permissiable.Companies like vlm.run and roboflow are using for commercial use as show by thier comments below. So i guess it can be used for commercial purposes.
It's interesting how some research in this space never took off as this needs government and regulation support over corporate profits. https://www.cs.cmu.edu/smartheadlight/
really great results although the results are so high i was trying a simple example of object detection and the performance was kind of poor in agentic frameworks. Need to see how this performs on other other tasks.
Really exciting results on paper. But truly interesting to see what data this has been trained on. There is a thin line between accuracy improvements and the data used from users. Hope the data used to train was obtained with consent from the creators