Update app.py

app.py

@@ -1,6 +1,7 @@
 import gradio as gr
 from transformers import AutoProcessor, AutoTokenizer, AutoImageProcessor, AutoModelForCausalLM, BlipForConditionalGeneration, VisionEncoderDecoderModel
 import torch
+import open_clip
 
 torch.hub.download_url_to_file('http://images.cocodataset.org/val2017/000000039769.jpg', 'cats.jpg')
 torch.hub.download_url_to_file('https://huggingface.co/datasets/nielsr/textcaps-sample/resolve/main/stop_sign.png', 'stop_sign.png')
@@ -22,6 +23,11 @@ vitgpt_processor = AutoImageProcessor.from_pretrained("nlpconnect/vit-gpt2-image
 vitgpt_model = VisionEncoderDecoderModel.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
 vitgpt_tokenizer = AutoTokenizer.from_pretrained("nlpconnect/vit-gpt2-image-captioning")
 
+coca_model, _, coca_transform = open_clip.create_model_and_transforms(
+    "coca_ViT-L-14",
+    pretrained="laion2B-s13B-b90k-mscoco-2014.pt"
+)
+
 device = "cuda" if torch.cuda.is_available() else "cpu"
 
 git_model_base.to(device)
@@ -29,6 +35,7 @@ blip_model_base.to(device)
 git_model_large.to(device)
 blip_model_large.to(device)
 vitgpt_model.to(device)
+coca_model.to(device)
 
 def generate_caption(processor, model, image, tokenizer=None):
     inputs = processor(images=image, return_tensors="pt").to(device)
@@ -43,6 +50,12 @@ def generate_caption(processor, model, image, tokenizer=None):
     return generated_caption
 
 
+def generate_caption_coca(model, transform, image):
+    im = transform(image).unsqueeze(0).to(device)
+    generated = model.generate(im, seq_len=20)
+    return open_clip.decode(generated[0].detach()).split("<end_of_text>")[0].replace("<start_of_text>", "")
+
+
 def generate_captions(image):
     caption_git_base = generate_caption(git_processor_base, git_model_base, image)
 
@@ -54,11 +67,13 @@ def generate_captions(image):
 
     caption_vitgpt = generate_caption(vitgpt_processor, vitgpt_model, image, vitgpt_tokenizer)
 
-    return caption_git_base, caption_git_large, caption_blip_base, caption_blip_large, caption_vitgpt
+    caption_coca = generate_caption_coca(coca_model, coca_transform, image)
+
+    return caption_git_base, caption_git_large, caption_blip_base, caption_blip_large, caption_vitgpt, caption_coca
 
    
 examples = [["cats.jpg"], ["stop_sign.png"], ["astronaut.jpg"]]
-outputs = [gr.outputs.Textbox(label="Caption generated by GIT-base"), gr.outputs.Textbox(label="Caption generated by GIT-large"), gr.outputs.Textbox(label="Caption generated by BLIP-base"), gr.outputs.Textbox(label="Caption generated by BLIP-large"), gr.outputs.Textbox(label="Caption generated by ViT+GPT-2")] 
+outputs = [gr.outputs.Textbox(label="Caption generated by GIT-base"), gr.outputs.Textbox(label="Caption generated by GIT-large"), gr.outputs.Textbox(label="Caption generated by BLIP-base"), gr.outputs.Textbox(label="Caption generated by BLIP-large"), gr.outputs.Textbox(label="Caption generated by ViT+GPT-2"), gr.outputs.Textbox(label="Caption generated by CoCa")] 
 
 title = "Interactive demo: comparing image captioning models"
 description = "Gradio Demo to compare GIT, BLIP and ViT+GPT2, 3 state-of-the-art vision+language models. To use it, simply upload your image and click 'submit', or click one of the examples to load them. Read more at the links below."