revantteotia / clip-training Goto Github PK
View Code? Open in Web Editor NEWCode to train CLIP model
License: MIT License
Code to train CLIP model
License: MIT License
Hello, @revantteotia !
I use this code to train CLIP from scratch using CC3M + CC12M dataset. But my loss curve seems to be weird as below.
As you can see, the loss would rise suddenly and drop faster than before. Do you meet the same problem?
PS: 1 epoch = 6700 step
Hi, First of all, Thanks for the great code base, it really helped me understand CLIP.
I was wondering if there is a COCO Image Retrieval
code available to test the capability of the extracted image features to find similar images? Writing the code would be easy but finding a standard code that everyone uses to test the Image Retrieval capacity of CLIP's image features would be really good.
The benchmarks at PapersWithCode each do it in their own way and I was wondering if there was a better and easier way to do it.
Thanks again
Have you tried fine-tune the clip model in mscoco?If you have tried, how the result is?Maybe, could you provide a fine-tune code? Thanks.
Hi.
Considering that the original CLIP source code has no training code, this repo can be a very valuable resource.
Would you consider adding a license? (I would suggest the MIT license, just because it's the license of CLIP).
Hi, Thanks for the great codebase for training CLIP from scratch. It really helps with understanding how it works.
I was wondering if there are any facilities to log and plot the training and evaluation losses? Something like Tensorboard or Weights&Biases?
Thanks again.
Hello, @revantteotia ~
I would like to use your great code to train my dataset, but I want to load ViT-B/32 Model. I see you have finished the code of VisualTransformer, but I have a little question about the parameter--vision_patch_size.
I read other ViT model code which set patch_size = 32 when the image resolution is 224, so I want to ask if the 32 is right in model_config.yaml if my image_resolution is 224.
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.