Add icon detect image size option

demo.ipynb

@@ -104,7 +104,7 @@
     "ocr_bbox_rslt, is_goal_filtered = check_ocr_box(image_path, display_img = False, output_bb_format='xyxy', goal_filtering=None, easyocr_args={'paragraph': False, 'text_threshold':0.9}, use_paddleocr=True)\n",
     "text, ocr_bbox = ocr_bbox_rslt\n",
     "\n",
-    "dino_labled_img, label_coordinates, parsed_content_list = get_som_labeled_img(image_path, som_model, BOX_TRESHOLD = BOX_TRESHOLD, output_coord_in_ratio=False, ocr_bbox=ocr_bbox,draw_bbox_config=draw_bbox_config, caption_model_processor=caption_model_processor, ocr_text=text,use_local_semantics=True, iou_threshold=0.1)\n",
+    "dino_labled_img, label_coordinates, parsed_content_list = get_som_labeled_img(image_path, som_model, BOX_TRESHOLD = BOX_TRESHOLD, output_coord_in_ratio=False, ocr_bbox=ocr_bbox,draw_bbox_config=draw_bbox_config, caption_model_processor=caption_model_processor, ocr_text=text,use_local_semantics=True, iou_threshold=0.1, imgsz=640)\n",
     "\n",
     "\n"
    ]

gradio_demo.py

@@ -61,7 +61,8 @@ def process(
     image_input,
     box_threshold,
     iou_threshold,
-    use_paddleocr
+    use_paddleocr,
+    imgsz
 ) -> Optional[Image.Image]:
 
     image_save_path = 'imgs/saved_image_demo.png'
@@ -71,7 +72,7 @@ def process(
     ocr_bbox_rslt, is_goal_filtered = check_ocr_box(image_save_path, display_img = False, output_bb_format='xyxy', goal_filtering=None, easyocr_args={'paragraph': False, 'text_threshold':0.9}, use_paddleocr=use_paddleocr)
     text, ocr_bbox = ocr_bbox_rslt
     # print('prompt:', prompt)
-    dino_labled_img, label_coordinates, parsed_content_list = get_som_labeled_img(image_save_path, yolo_model, BOX_TRESHOLD = box_threshold, output_coord_in_ratio=True, ocr_bbox=ocr_bbox,draw_bbox_config=draw_bbox_config, caption_model_processor=caption_model_processor, ocr_text=text,iou_threshold=iou_threshold)
+    dino_labled_img, label_coordinates, parsed_content_list = get_som_labeled_img(image_save_path, yolo_model, BOX_TRESHOLD = box_threshold, output_coord_in_ratio=True, ocr_bbox=ocr_bbox,draw_bbox_config=draw_bbox_config, caption_model_processor=caption_model_processor, ocr_text=text,iou_threshold=iou_threshold, imgsz=imgsz)
     image = Image.open(io.BytesIO(base64.b64decode(dino_labled_img)))
     print('finish processing')
     parsed_content_list = '\n'.join(parsed_content_list)
@@ -93,6 +94,8 @@ with gr.Blocks() as demo:
                 label='IOU Threshold', minimum=0.01, maximum=1.0, step=0.01, value=0.1)
             use_paddleocr_component = gr.Checkbox(
                 label='Use PaddleOCR', value=True)
+            imgsz_component = gr.Slider(
+                label='Icon Detect Image Size', minimum=640, maximum=1920, step=32, value=640)
             submit_button_component = gr.Button(
                 value='Submit', variant='primary')
         with gr.Column():
@@ -105,7 +108,8 @@ with gr.Blocks() as demo:
             image_input_component,
             box_threshold_component,
             iou_threshold_component,
-            use_paddleocr_component
+            use_paddleocr_component,
+            imgsz_component
         ],
         outputs=[image_output_component, text_output_component]
     )

utils.py

@@ -280,7 +280,7 @@ def predict(model, image, caption, box_threshold, text_threshold):
     return boxes, logits, phrases
 
 
-def predict_yolo(model, image_path, box_threshold):
+def predict_yolo(model, image_path, box_threshold, imgsz):
     """ Use huggingface model to replace the original model
     """
     # model = model['model']
@@ -288,6 +288,7 @@ def predict_yolo(model, image_path, box_threshold):
     result = model.predict(
     source=image_path,
     conf=box_threshold,
+    imgsz=imgsz
     # iou=0.5, # default 0.7
     )
     boxes = result[0].boxes.xyxy#.tolist() # in pixel space
@@ -297,7 +298,7 @@ def predict_yolo(model, image_path, box_threshold):
     return boxes, conf, phrases
 
 
-def get_som_labeled_img(img_path, model=None, BOX_TRESHOLD = 0.01, output_coord_in_ratio=False, ocr_bbox=None, text_scale=0.4, text_padding=5, draw_bbox_config=None, caption_model_processor=None, ocr_text=[], use_local_semantics=True, iou_threshold=0.9,prompt=None):
+def get_som_labeled_img(img_path, model=None, BOX_TRESHOLD = 0.01, output_coord_in_ratio=False, ocr_bbox=None, text_scale=0.4, text_padding=5, draw_bbox_config=None, caption_model_processor=None, ocr_text=[], use_local_semantics=True, iou_threshold=0.9,prompt=None,imgsz=640):
     """ ocr_bbox: list of xyxy format bbox
     """
     TEXT_PROMPT = "clickable buttons on the screen"
@@ -309,7 +310,7 @@ def get_som_labeled_img(img_path, model=None, BOX_TRESHOLD = 0.01, output_coord_
     if False: # TODO
         xyxy, logits, phrases = predict(model=model, image=image_source, caption=TEXT_PROMPT, box_threshold=BOX_TRESHOLD, text_threshold=TEXT_TRESHOLD)
     else:
-        xyxy, logits, phrases = predict_yolo(model=model, image_path=img_path, box_threshold=BOX_TRESHOLD)
+        xyxy, logits, phrases = predict_yolo(model=model, image_path=img_path, box_threshold=BOX_TRESHOLD, imgsz=imgsz)
     xyxy = xyxy / torch.Tensor([w, h, w, h]).to(xyxy.device)
     image_source = np.asarray(image_source)
     phrases = [str(i) for i in range(len(phrases))]