-
Thank you for your great work! In the paper, you described LM over SuperGLUE being evaluated for zero-shot learning. Could you please share the code for that evaluation?
-
Hello,
Could you please let me know where exactly is the "novelty detection" part of the paper "Zero-Shot Learning Through Cross-Modal Transfer" in your code.
-
Planning new features in `autogen`. Suggestions are welcome.
-
Good day! First, I'd like to say great work on this!
As I was trying to reproduce the results found [here](https://github.com/microsoft/RegionCLIP/blob/main/docs/MODEL_ZOO.md), I'd like to focus on…
-
### Search before asking
- [X] I have searched the YOLOv8 [issues](https://github.com/ultralytics/ultralytics/issues) and [discussions](https://github.com/ultralytics/ultralytics/discussions) and fou…
-
![image](https://user-images.githubusercontent.com/46675408/150263175-a405612c-b130-4a53-ab18-88b8eb199107.png)
[arxiv](https://arxiv.org/pdf/2112.12750v1.pdf), [code](https://github.com/facebookrese…
-
Could you share the datasets and code for cold-start recommendation (it seems you only provide code and data for zero-shot learning)? It will be better if you can provide the training and testing sets…
-
Why does `unsenn` appear in the loss function even though it is not used during the training phase to mitigate bias?
-
Dear author, thank you for your code and paper. I noticed that in your paper ‘Multi-Head Self-Attention via Vision Transformer for Zero-Shot Learning’, the attributes dimension of the CUB (102 -> 312)…
-
Great work and thanks for the codebase!
I want to know the exact detailed of LoRA fine-tuning as mentioned in Table 6 of the main paper.
Also if you could point-out to the bash script to reproduce…