Update app.py

app.py

@@ -1,89 +1,44 @@
+import torch 
+import re 
 import gradio as gr
-from transformers import AutoProcessor, AutoTokenizer, AutoImageProcessor, AutoModelForCausalLM, BlipForConditionalGeneration, Blip2ForConditionalGeneration, VisionEncoderDecoderModel
-import torch
-import open_clip
-
-from huggingface_hub import hf_hub_download
-
-
-git_processor_large_coco = AutoProcessor.from_pretrained("microsoft/git-large-coco")
-git_model_large_coco = AutoModelForCausalLM.from_pretrained("microsoft/git-large-coco")
-
-git_processor_large_textcaps = AutoProcessor.from_pretrained("microsoft/git-large-r-textcaps")
-git_model_large_textcaps = AutoModelForCausalLM.from_pretrained("microsoft/git-large-r-textcaps")
-
-
-blip_processor_large = AutoProcessor.from_pretrained("Salesforce/blip-image-captioning-large")
-blip_model_large = BlipForConditionalGeneration.from_pretrained("Salesforce/blip-image-captioning-large")
-
-blip2_processor_8_bit = AutoProcessor.from_pretrained("Salesforce/blip2-opt-6.7b")
-blip2_model_8_bit = Blip2ForConditionalGeneration.from_pretrained("Salesforce/blip2-opt-6.7b", device_map="auto", load_in_8bit=True)
-
-coca_model, _, coca_transform = open_clip.create_model_and_transforms(
-  model_name="coca_ViT-L-14",
-  pretrained="mscoco_finetuned_laion2B-s13B-b90k"
-)
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-
-git_model_large_coco.to(device)
-git_model_large_textcaps.to(device)
-blip_model_large.to(device)
-coca_model.to(device)
-
-def generate_caption(processor, model, image, tokenizer=None, use_float_16=False):
-    inputs = processor(images=image, return_tensors="pt").to(device)
-
-    if use_float_16:
-        inputs = inputs.to(torch.float16)
-    
-    generated_ids = model.generate(pixel_values=inputs.pixel_values, max_length=50)
-
-    if tokenizer is not None:
-        generated_caption = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
-    else:
-        generated_caption = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
-   
-    return generated_caption
-
-
-def generate_caption_coca(model, transform, image):
-    im = transform(image).unsqueeze(0).to(device)
-    with torch.no_grad(), torch.cuda.amp.autocast():
-        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_large_coco = generate_caption(git_processor_large_coco, git_model_large_coco, image)
-
-    caption_git_large_textcaps = generate_caption(git_processor_large_textcaps, git_model_large_textcaps, image)
-
-    caption_blip_large = generate_caption(blip_processor_large, blip_model_large, image)
-
-    caption_coca = generate_caption_coca(coca_model, coca_transform, image)
-
-    caption_blip2_8_bit = generate_caption(blip2_processor_8_bit, blip2_model_8_bit, image, use_float_16=True).strip()
-
-    return caption_git_large_coco, caption_git_large_textcaps, caption_blip_large, caption_coca, caption_blip2_8_bit
-
-   
-
-examples = [["Image1.jpg"], ["Image2.jpg"], ["Image3.jpg"]]
-outputs = [gr.outputs.Textbox(label="Caption generated - 1"), gr.outputs.Textbox(label="Caption generated - 2"), gr.outputs.Textbox(label="Caption generated -3"), gr.outputs.Textbox(label="Caption generated - 4"), gr.outputs.Textbox(label="Caption generated - 5")] 
-
-
-title = "Interactive demo: comparing image captioning models"
-description = "Image Caption Generator by Sravanth Kurmala"
-article = "Assignment for Listed Inc"
-
-interface = gr.Interface(fn=generate_captions, 
-                         inputs=gr.inputs.Image(type="pil"),
-                         outputs=outputs,
-                         examples=examples, 
-                         title=title,
-                         description=description,
-                         article=article, 
-                         enable_queue=True)
+from transformers import AutoTokenizer, ViTFeatureExtractor, VisionEncoderDecoderModel 
+
+device='cpu'
+encoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
+decoder_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
+model_checkpoint = "nlpconnect/vit-gpt2-image-captioning"
+feature_extractor = ViTFeatureExtractor.from_pretrained(encoder_checkpoint)
+tokenizer = AutoTokenizer.from_pretrained(decoder_checkpoint)
+model = VisionEncoderDecoderModel.from_pretrained(model_checkpoint).to(device)
+
+
+def predict(image,max_length=64, num_beams=4):
+  image = image.convert('RGB')
+  image = feature_extractor(image, return_tensors="pt").pixel_values.to(device)
+  clean_text = lambda x: x.replace('<|endoftext|>','').split('\n')[0]
+  caption_ids = model.generate(image, max_length = max_length)[0]
+  caption_text = clean_text(tokenizer.decode(caption_ids))
+  return caption_text 
+
+
+
+input = gr.inputs.Image(label="Upload your Image", type = 'pil', optional=True)
+output = gr.outputs.Textbox(type="auto",label="Captions")
+examples = [f"example{i}.jpg" for i in range(1,5)]
+
+description= "Image caption Generator"
+title = "Image Captioning 🖼️"
+
+article = "Created By : Sravanth Kurmala"
+
+interface = gr.Interface(
+        fn=predict,
+        inputs = input,
+        theme="grass",
+        outputs=output,
+        examples = examples,
+        title=title,
+        description=description,
+        article = article,
+    )
 interface.launch(debug=True)