Upload LLaVA-Next-3D/data_precessing/box_visualization.py with huggingface_hub

LLaVA-Next-3D/data_precessing/box_visualization.py

@@ -0,0 +1,54 @@
+from PIL import Image, ImageDraw, ImageFont
+import json
+import os
+from diffusers.utils import make_image_grid
+import pdb
+
+def draw_bbox_on_image(path, bbox_normalized):
+    # 打开图像文件
+    img = Image.open(path)
+    width, height = img.size
+    x1 = int(bbox_normalized[0] * width)
+    y1 = int(bbox_normalized[1] * height)
+    x2 = int(bbox_normalized[2] * width)
+    y2 = int(bbox_normalized[3] * height)
+
+    draw = ImageDraw.Draw(img)
+    draw.rectangle([(x1, y1), (x2, y2)], outline='red', width=2)
+    return img
+
+result_file = 'results/scanrefer/prompt-llavanext-qwen-uniform-16bs-scanrefer-wosam-uniform.jsonl'
+visual_num = 1000
+
+with open(result_file, 'r') as f:
+    i = 0
+    for line in f:
+        i += 1
+        if i >= visual_num:
+            break
+        
+        item = json.loads(line)
+        output_dir = result_file.replace('.jsonl', '')
+        os.makedirs(output_dir, exist_ok=True)
+
+        pred_response = item['pred_response']
+        description = item['prompt'].split('\n')[-1]
+        sample_id = str(item['sample_id'])
+        sample_dir = os.path.join(output_dir, sample_id)
+
+        image_list = []
+        for path, box in pred_response.items():
+            try:
+                drawn_image = draw_bbox_on_image(path, box)
+                image_list.append(drawn_image)
+                # drawn_image.save(os.path.join(sample_dir, path.split('/')[-1]))
+            except:
+                continue
+        
+        image_num = min(5, len(image_list))
+        image_list = image_list[:image_num]
+        concat_image = make_image_grid(image_list, 1, image_num)
+        draw = ImageDraw.Draw(concat_image)
+        font = ImageFont.load_default(size=80)
+        draw.text((10, 10), description, fill='green', font=font)
+        concat_image.save(os.path.join(sample_dir+'.jpg'))