Update app.py

app.py

@@ -5,72 +5,72 @@ from transformers import CLIPProcessor, CLIPModel, YolosImageProcessor, YolosFor
 import torch
 
 st.title("CLIP & CROP")
-st.markdown("**Extract sections of images from your image by using OpenAI's CLIP and Facebooks Detr implemented on HuggingFace Transformers, if the similarity score is not so much, then please consider the prediction to be void.**")
+# st.markdown("**Extract sections of images from your image by using OpenAI's CLIP and Facebooks Detr implemented on HuggingFace Transformers, if the similarity score is not so much, then please consider the prediction to be void.**")
 
-IMAGE_INPUT = st.file_uploader(type=["jpg", "png"], label="Input image")
-TEXT_INPUT = st.text_input(label="Description for section to extracted")
-NUMBER_INPUT = st.number_input(value=0.96, label="Threshold percentage score")
+# IMAGE_INPUT = st.file_uploader(type=["jpg", "png"], label="Input image")
+# TEXT_INPUT = st.text_input(label="Description for section to extracted")
+# NUMBER_INPUT = st.number_input(value=0.96, label="Threshold percentage score")
 
 
-with st.spinner("Models are loading"):
-    feature_extractor = YolosImageProcessor.from_pretrained("hustvl/yolos-tiny")
-    dmodel = YolosForObjectDetection.from_pretrained('hustvl/yolos-tiny')
+# with st.spinner("Models are loading"):
+#     feature_extractor = YolosImageProcessor.from_pretrained("hustvl/yolos-tiny")
+#     dmodel = YolosForObjectDetection.from_pretrained('hustvl/yolos-tiny')
 
-    model = CLIPModel.from_pretrained("openai/clip-vit-base-patch16")
-    processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch16")
+#     model = CLIPModel.from_pretrained("openai/clip-vit-base-patch16")
+#     processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch16")
  
-SUBMIT_BUTTON = st.button("SUBMIT")
+# SUBMIT_BUTTON = st.button("SUBMIT")
 
-def extract_image(image, text, prob, num=1):
+# def extract_image(image, text, prob, num=1):
     
-    inputs = feature_extractor(images=image, return_tensors="pt")
-    outputs = dmodel(**inputs)
+#     inputs = feature_extractor(images=image, return_tensors="pt")
+#     outputs = dmodel(**inputs)
     
-    # model predicts bounding boxes and corresponding COCO classes
-    logits = outputs.logits
-    bboxes = outputs.pred_boxes
-    probas = outputs.logits.softmax(-1)[0, :, :-1] #removing no class as detr maps 
+#     # model predicts bounding boxes and corresponding COCO classes
+#     logits = outputs.logits
+#     bboxes = outputs.pred_boxes
+#     probas = outputs.logits.softmax(-1)[0, :, :-1] #removing no class as detr maps 
     
-    keep = probas.max(-1).values > prob
-    outs = feature_extractor.post_process(outputs, torch.tensor(image.size[::-1]).unsqueeze(0))
-    bboxes_scaled = outs[0]['boxes'][keep].detach().numpy()
-    labels = outs[0]['labels'][keep].detach().numpy()
-    scores = outs[0]['scores'][keep].detach().numpy()
+#     keep = probas.max(-1).values > prob
+#     outs = feature_extractor.post_process(outputs, torch.tensor(image.size[::-1]).unsqueeze(0))
+#     bboxes_scaled = outs[0]['boxes'][keep].detach().numpy()
+#     labels = outs[0]['labels'][keep].detach().numpy()
+#     scores = outs[0]['scores'][keep].detach().numpy()
     
-    images_list = []
-    for i,j in enumerate(bboxes_scaled):
+#     images_list = []
+#     for i,j in enumerate(bboxes_scaled):
       
-      xmin = int(j[0])
-      ymin = int(j[1])
-      xmax = int(j[2])
-      ymax = int(j[3])
+#       xmin = int(j[0])
+#       ymin = int(j[1])
+#       xmax = int(j[2])
+#       ymax = int(j[3])
     
-      im_arr = np.array(image)
-      roi = im_arr[ymin:ymax, xmin:xmax]
-      roi_im = Image.fromarray(roi)
+#       im_arr = np.array(image)
+#       roi = im_arr[ymin:ymax, xmin:xmax]
+#       roi_im = Image.fromarray(roi)
     
-      images_list.append(roi_im)
+#       images_list.append(roi_im)
     
-    inpu = processor(text = [text], images=images_list , return_tensors="pt", padding=True)
-    output = model(**inpu)
-    logits_per_image = output.logits_per_text
-    probs = logits_per_image.softmax(-1)
-    l_idx = np.argsort(probs[-1].detach().numpy())[::-1][0:num]
+#     inpu = processor(text = [text], images=images_list , return_tensors="pt", padding=True)
+#     output = model(**inpu)
+#     logits_per_image = output.logits_per_text
+#     probs = logits_per_image.softmax(-1)
+#     l_idx = np.argsort(probs[-1].detach().numpy())[::-1][0:num]
     
-    final_ims = []
-    for i,j in enumerate(images_list):
-      json_dict = {}
-      if i in l_idx:
-        json_dict['image'] = images_list[i]
-        json_dict['score'] = probs[-1].detach().numpy()[i]
+#     final_ims = []
+#     for i,j in enumerate(images_list):
+#       json_dict = {}
+#       if i in l_idx:
+#         json_dict['image'] = images_list[i]
+#         json_dict['score'] = probs[-1].detach().numpy()[i]
     
-        final_ims.append(json_dict)
+#         final_ims.append(json_dict)
     
-    fi = sorted(final_ims, key=lambda item: item.get("score"), reverse=True)
-    return fi[0]['image'], fi[0]['score']
+#     fi = sorted(final_ims, key=lambda item: item.get("score"), reverse=True)
+#     return fi[0]['image'], fi[0]['score']
 
-if SUBMIT_BUTTON:
-    imageOutput, scoreOutput = extract(IMAGE_INPUT, TEXT_INPUT, NUMBER_INPUT)
-    st.image(imageOutput, caption="Cropped Image")
-    st.markdown("*Confidence Score:*")
-    st.success(scoreOutput)
+# if SUBMIT_BUTTON:
+#     imageOutput, scoreOutput = extract(IMAGE_INPUT, TEXT_INPUT, NUMBER_INPUT)
+#     st.image(imageOutput, caption="Cropped Image")
+#     st.markdown("*Confidence Score:*")
+#     st.success(scoreOutput)