NoneType error
FanLu1994 opened this issue · comments
I'm using tensorflow1.10 on ubuntu18.04. how can i fix this problem
Traceback (most recent call last):
File "./tools/demo.py", line 114, in
net = get_network(args.demo_net)
File "/home/fanlu/Faster-RCNN_TF/tools/../lib/networks/factory.py", line 28, in get_network
return networks.VGGnet_test()
File "/home/fanlu/Faster-RCNN_TF/tools/../lib/networks/VGGnet_test.py", line 16, in init
self.setup()
File "/home/fanlu/Faster-RCNN_TF/tools/../lib/networks/VGGnet_test.py", line 57, in setup
.fc(4096, name='fc6')
File "/home/fanlu/Faster-RCNN_TF/tools/../lib/networks/network.py", line 25, in layer_decorated
layer_output = op(self, layer_input, *args, **kwargs)
File "/home/fanlu/Faster-RCNN_TF/tools/../lib/networks/network.py", line 246, in fc
feed_in, dim = (input, int(input_shape[-1]))
TypeError: int returned non-int (type NoneType)
Hi, I also encountered the same issue. Do you solve this problem?
@qingping95 sorry, i didn't solve it. it's probably caused by TF version. I’m using faster-rcnn.pytorch. https://github.com/jwyang/faster-rcnn.pytorch
@CodeForWuyu Yeah, You are right! I just now solved it by changing my tensorflow version to 1.8.1. Thank you for your reply.
I was able to make it work for TF 1.10.1 after making the following changes:
add the following lines to lib/roi_pooling_layer/roi_pooling_op.cc
#include "tensorflow/core/framework/shape_inference.h"
...
.Output("argmax: int32")
.SetShapeFn([](::tensorflow::shape_inference::InferenceContext* c) {
//https://github.com/tensorflow/.../core/framework/shape_inference.h
int pooled_height;
int pooled_width;
c->GetAttr("pooled_height", &pooled_height);
c->GetAttr("pooled_width", &pooled_width);
auto pooled_height_h = c->MakeDim(pooled_height);
auto pooled_width_h = c->MakeDim(pooled_width);
auto output_shape = c->MakeShape({ c->Dim(c->input(1), 0), pooled_height_h, pooled_width_h, c->Dim(c->input(0), 3) });
c->set_output(0, output_shape);
return Status::OK();
});
You can also find the changes at https://github.com/saizhang12/Faster-RCNN_TF/blob/master/lib/roi_pooling_layer/roi_pooling_op.cc
Amazing !