📝 Investigating the Limitation of CLIP Models: The Worst-Performing Categories 🔭🧠

"Proposes the Class-wise Matching Margin (\cmm) to measure the inference confusion and find the worst-performing categories of CLIP models without any manual prompt engineering, laborious optimization, or access to labeled validation data." [gal30b+] 🤖 #CV #LG

⚙️ https://github.com/openai/CLIP
🔗 https://arxiv.org/abs/2310.03324v1 #arxiv

https://creative.ai/system/media_attachments/files/111/190/292/457/863/039/original/5685dbbeaab8ee65.jpg

https://creative.ai/system/media_attachments/files/111/190/292/519/546/649/original/9e41141bf3007451.jpg

https://creative.ai/system/media_attachments/files/111/190/292/579/494/295/original/0f492461e37f9023.jpg

https://creative.ai/system/media_attachments/files/111/190/292/647/778/170/original/643da9649d77e274.jpg