init

llava/conversation.py

@@ -133,17 +133,17 @@ class Conversation:
             image = image.resize((336, 336))
         else:
             raise ValueError(f"Invalid image_process_mode: {image_process_mode}")
-        if max(image.size) > max_len:
-            max_hw, min_hw = max(image.size), min(image.size)
-            aspect_ratio = max_hw / min_hw
-            shortest_edge = int(min(max_len / aspect_ratio, min_len, min_hw))
-            longest_edge = int(shortest_edge * aspect_ratio)
-            W, H = image.size
-            if H > W:
-                H, W = longest_edge, shortest_edge
-            else:
-                H, W = shortest_edge, longest_edge
-            image = image.resize((W, H))
+        # if max(image.size) > max_len:
+        #     max_hw, min_hw = max(image.size), min(image.size)
+        #     aspect_ratio = max_hw / min_hw
+        #     shortest_edge = int(min(max_len / aspect_ratio, min_len, min_hw))
+        #     longest_edge = int(shortest_edge * aspect_ratio)
+        #     W, H = image.size
+        #     if H > W:
+        #         H, W = longest_edge, shortest_edge
+        #     else:
+        #         H, W = shortest_edge, longest_edge
+        #     image = image.resize((W, H))
         if return_pil:
             return image
         else:

llava/serve/gradio_web_server.py

@@ -61,7 +61,7 @@ def resize_image(image, default_width=(1344, 896), request_width=None):
 from PIL import Image, ImageDraw
 
 
-def draw_circle_on_image(image, x, y, radius=10, color=(255, 0, 0)):
+def draw_circle_on_image(image, x, y, radius=20, color=(255, 0, 0)):
     """
     在给定的图片上绘制一个红色圆圈，并返回新的图片。如果 x, y 坐标不在图片范围内，
     并且 y 超出了图片高度，则尝试将 y 减去 224；如果调整后的 y 仍然超出范围，则返回原图。
@@ -382,6 +382,7 @@ def http_bot(state, model_selector, temperature, top_p, max_new_tokens, request:
 
     if len(all_images) > 0:
         # 假设我们对第一张图片进行 resize 并展示
+
         resized_image = draw_circle_on_image(resize_image(all_images[0]),original_coord[0],original_coord[1])
         # state.append_message(state.roles[1], ("", resized_image,"Default"))
         yield (state, state.to_gradio_chatbot(resized_image)) + (enable_btn,) * 5
@@ -425,8 +426,8 @@ block_css = """
 }
 
 #chatbot img {
-    max-width: 100%;    /* 宽图片根据宽度调整 */
-    max-height: 100vh;  /* 高图片根据视口高度调整 */
+    max-width: 80%;    /* 宽图片根据宽度调整 */
+    max-height: 80vh;  /* 高图片根据视口高度调整 */
     width: auto;        /* 保持宽度自适应 */
     height: auto;       /* 保持高度自适应 */
     object-fit: contain; /* 保持图片宽高比，不失真 */
@@ -444,13 +445,14 @@ def build_demo(embed_mode, cur_dir=None, concurrency_count=1):
 
         with gr.Row():
             with gr.Column(scale=3):
-                with gr.Row(elem_id="model_selector_row"):
-                    model_selector = gr.Dropdown(
-                        choices=models,
-                        value=models[0] if len(models) > 0 else "",
-                        interactive=True,
-                        show_label=False,
-                        container=False)
+                # with gr.Row(elem_id="model_selector_row"):
+                #     model_selector = gr.Dropdown(
+                #         choices=models,
+                #         value=models[0] if len(models) > 0 else "",
+                #         interactive=True,
+                #         show_label=False,
+                #         container=False)
+                model_selector="llava-v1.5-UGround_v1"
 
                 imagebox = gr.Image(type="pil")
                 image_process_mode = gr.Radio(
@@ -464,11 +466,14 @@ def build_demo(embed_mode, cur_dir=None, concurrency_count=1):
                     [f"{cur_dir}/amazon.jpg",f"Search bar at the top of the page"],
                     # [f"{cur_dir}/examples/waterview.jpg", "What are the things I should be cautious about when I visit here?"],
                 ], inputs=[imagebox, textbox])
-
-                with gr.Accordion("Parameters", open=False) as parameter_row:
-                    temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.2, step=0.1, interactive=True, label="Temperature",)
-                    top_p = gr.Slider(minimum=0.0, maximum=1.0, value=0.7, step=0.1, interactive=True, label="Top P",)
-                    max_output_tokens = gr.Slider(minimum=0, maximum=1024, value=512, step=64, interactive=True, label="Max output tokens",)
+                temperature=0
+                top_p=0.7
+                max_output_tokens=16384
+                #
+                # with gr.Accordion("Parameters", open=False) as parameter_row:
+                #     temperature = gr.Slider(minimum=0.0, maximum=1.0, value=0.2, step=0.1, interactive=True, label="Temperature",)
+                #     top_p = gr.Slider(minimum=0.0, maximum=1.0, value=0.7, step=0.1, interactive=True, label="Top P",)
+                #     max_output_tokens = gr.Slider(minimum=0, maximum=1024, value=512, step=64, interactive=True, label="Max output tokens",)
 
             with gr.Column(scale=8):
                 chatbot = gr.Chatbot(