Fix RMBG-2.0: use briaai/RMBG-2.0 (official) with standard load pattern

app.py

@@ -249,11 +249,11 @@ def load_triposg():
 
     try:
         from transformers import AutoModelForImageSegmentation
-        # device_map loads weights directly to GPU, avoiding the meta tensor
-        # intermediate that causes "Tensor.item() cannot be called on meta tensors"
         _rmbg_net = AutoModelForImageSegmentation.from_pretrained(
-            "1038lab/RMBG-2.0", trust_remote_code=True, device_map=DEVICE
+            "briaai/RMBG-2.0", trust_remote_code=True
         )
+        torch.set_float32_matmul_precision("high")
+        _rmbg_net.to(DEVICE)
         _rmbg_net.eval()
         _rmbg_version = "2.0"
         print("[load_triposg] TripoSG + RMBG-2.0 loaded.")