Closed allendred closed 5 years ago
Hi,
first please make sure that u r using pytorch 0.4.0.
then go to your /usr/lib/python3.6/site-packages/torch/onnx or your virtual env lib/python3.6/site-packages/torch/onnx
replace utils.py line 362 with attrs = {k: n[k] for k in n.attributeNames() if k != 'align_corners'}
replace symbolic.py line 461-466 with
def upsample_bilinear2d(g, input, output_size): input_shape_str = re.findall(r'[^()!]+', str(input))[2].split(',') w_scale = float(output_size[-1]) / float(input_shape_str[-1]) h_scale = float(output_size[-2]) / float(input_shape_str[-2]) return g.op("Upsample", input, channel_s=input_shape_str[1], width_scale_f=w_scale, height_scale_f=h_scale, mode_s="bilinear")
I will close this issue, u can reopen it if there is any update
Something wrong about "align_corners" in function "upsample".
Detail: TypeError: upsample_bilinear2d() got an unexpected keyword argument 'align_corners' (occurred when translating upsample_bilinear2d)