Update README.md

README.md

@@ -6,3 +6,23 @@ pipeline_tag: zero-shot-image-classification
 license: mit
 ---
 # Model card for open_clip_quilt1m_ft_cy_1
+
+This model is finetuned based on the Quilt-1M VIT-B-32 model using Chaoyang Dataset. 
+The training csv file is : `/dataset/chaoyang/chaoyang_train_multi_annos.csv`  
+For this model, I insert the multi-labels into the prompts. Since the "normal" and "serrated" are the adj, so I add the nouns for better expression.  
+The Paired Text used for training as listed below:  
+
+normal: "normal histology"
+serrated: "serrated polyps"
+adenocarcinomas: "adenocarcinomas"
+adenomas: "adenomas"
+
+So for different combinations, replacing the labels with corresponding words.   
+
+"normal histology, normal histology, normal histology"  
+"serrated polyps, serrated polyps, serrated polyps"  
+"adenocarcinomas, adenocarcinomas, adenocarcinomas"  
+"adenomas, adenomas, adenomas"  
+"adenomas, normal histology, adenomas"   
+
+The model is finetuned with Chaoyang Dataset with 64 epochs, but I choose the **32th** checkpoint as the final model according to the plot of loss. I.e., the loss began kept stable.