Fix screenshot case (#2)

- Fix screenshot case (3d2275a48a81590a56b66a0641e6f97cf1f8e03f)


Co-authored-by: Joya Chen <[email protected]>

infer.py

@@ -1,4 +1,5 @@
 # Copyright (2025) [Seed-VL-Cookbook] Bytedance Seed
+import cv2
 import json
 import time
 import math
@@ -9,8 +10,8 @@ import torch
 import decord
 import numpy as np
 from PIL import Image, ImageSequence
-from torchvision.io import read_image, encode_jpeg, ImageReadMode
-from torchvision.transforms.functional import resize
+from torchvision.io import read_image, encode_jpeg
+from torchvision.transforms.functional import resize, pil_to_tensor
 from torchvision.transforms import InterpolationMode
 
 
@@ -210,7 +211,18 @@ class SeedVLInfer:
                         },
                     })
             else:
-                image = read_image(path, ImageReadMode.RGB)
+                try:
+                    image = read_image(path, "RGB")
+                except:
+                    try:
+                        image = pil_to_tensor(Image.open(path).convert('RGB'))
+                    except:
+                        image = torch.from_numpy(
+                            cv2.cvtColor(
+                                cv2.imread(path), 
+                                cv2.COLOR_BGR2RGB
+                            )
+                        ).permute(2, 0, 1)
                 if online and path.endswith('.webp'):
                     streaming_timestamp = i
                 if streaming_timestamp is not None: