ONNXRuntimeError

[JING-SHAO]

When reconstructing a 4D mesh using the data in 'examples', get an error:
'onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Node (/decoder/Split_33) Op (Split) [ShapeInferenceError] Cannot parse data from external tensors. Please load external data into raw data for tensor: /decoder/Constant_1066_output_0'

[JING-SHAO]

Traceback (most recent call last):
File "/V2M4/main.py", line 480, in
_, rmbg_image_rgba, rmbg_image = TrellisImageTo3DPipeline.preprocess_image(image, return_all_rbga=True)
File "/V2M4/v2m4_trellis/pipelines/trellis_image_to_3d.py", line 106, in preprocess_image
rembg_session = rembg.new_session('birefnet-massive')
File "/anaconda3/envs/v2m4/lib/python3.10/site-packages/rembg/session_factory.py", line 48, in new_session
return session_class(model_name, sess_opts, *args, **kwargs)
File "/anaconda3/envs/v2m4/lib/python3.10/site-packages/rembg/sessions/base.py", line 34, in init
self.inner_session = ort.InferenceSession(
File "/anaconda3/envs/v2m4/lib/python3.10/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 473, in init
self._create_inference_session(providers, provider_options, disabled_optimizers)
File "/anaconda3/envs/v2m4/lib/python3.10/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 572, in _create_inference_session
sess.initialize_session(providers, provider_options, disabled_optimizers)
onnxruntime.capi.onnxruntime_pybind11_state.Fail: [ONNXRuntimeError] : 1 : FAIL : Node (/decoder/Split_33) Op (Split) [ShapeInferenceError] Cannot parse data from external tensors. Please load external data into raw data for tensor: /decoder/Constant_1066_output_0

[WindVChen]

Hi @JING-SHAO ,

It seems that the rembg package is not installed correctly. Please try reinstalling it and refer to its source page
for detailed installation instructions.

Hope this helps.

[JING-SHAO]

Hi Chen, Thanks for your reply. I have followed its source page to install the *rembg*, the problem is still unresolved. There is a github issue of *birefnet*: ZhengPeng7/BiRefNet#139. The author mentioned it may because of the version of pytorch since the torch's version I use is 2.6. However, after installing the torch=2.4, another error occurred(please see the image). It seems because the function 'torch.load' no longer support the torch under 2.6. I am not sure whether you could run the repo successfully now and if you have any suggestions about this problems, please let me know. Thank you~ [image: image.png] [image: b7262334913c200bf2c8ad8477980ca3.png] WindVChen ***@***.***> 于2025年9月30日周二 18:10写道：

…

*WindVChen* left a comment (WindVChen/V2M4#5) <#5 (comment)> Hi @JING-SHAO <https://github.com/JING-SHAO> , It seems that the rembg package is not installed correctly. Please try reinstalling it and refer to its source page <https://github.com/danielgatis/rembg> for detailed installation instructions. Hope this helps. — Reply to this email directly, view it on GitHub <#5 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/A3WBPMWU53476QEILQOTY5T3VJJJDAVCNFSM6AAAAACH4DCEBSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGNJRGEYTCNBZG4> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[JING-SHAO]

Hi Chen, I test 'sam', 'u2net' in rembg, and they all can run successfully, except for 'birefnet'. So could you please give me the model cache of BirefNet you used? I would appreciate it. Thank you very much. Jing Shao ***@***.***> 于2025年10月3日周五 18:23写道：

…

Hi Chen, Thanks for your reply. I have followed its source page to install the *rembg*, the problem is still unresolved. There is a github issue of *birefnet*: ZhengPeng7/BiRefNet#139. The author mentioned it may because of the version of pytorch since the torch's version I use is 2.6. However, after installing the torch=2.4, another error occurred(please see the image). It seems because the function 'torch.load' no longer support the torch under 2.6. I am not sure whether you could run the repo successfully now and if you have any suggestions about this problems, please let me know. Thank you~ [image: image.png] [image: b7262334913c200bf2c8ad8477980ca3.png] WindVChen ***@***.***> 于2025年9月30日周二 18:10写道： > *WindVChen* left a comment (WindVChen/V2M4#5) > <#5 (comment)> > > Hi @JING-SHAO <https://github.com/JING-SHAO> , > > It seems that the rembg package is not installed correctly. Please try > reinstalling it and refer to its source page > <https://github.com/danielgatis/rembg> > for detailed installation instructions. > > Hope this helps. > > — > Reply to this email directly, view it on GitHub > <#5 (comment)>, or > unsubscribe > <https://github.com/notifications/unsubscribe-auth/A3WBPMWU53476QEILQOTY5T3VJJJDAVCNFSM6AAAAACH4DCEBSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGNJRGEYTCNBZG4> > . > You are receiving this because you were mentioned.Message ID: > ***@***.***> >

[WindVChen]

Hi @JING-SHAO ,

The model (birefnet-massive) weight seems directly accessible from rembg source website, as here. The weight is in ONNX format, and I used the pretrained weight of that format.