bcnn_finetuning.py run error

[Hedlen]

When I run the bcnn_finetuning.py,and the following error occurs.My category is 134 classes.

InvalidArgumentError (see above for traceback): logits and labels must be same size: logits_size=[4,134] labels_size=[16,134] [[Node: SoftmaxCrossEntropyWithLogitsSoftmaxCrossEntropyWithLogits[T=DT_FLOAT, _device="/job:loc alhost/replica:0/task:0/gpu:0"](Reshape_2, Reshape_3)]]
logits_size and labels_size are not equal.
I don't know why this happens. I'm running exactly as you follow your code.
Please help me solve this problem.
Thank you very much!

[Hedlen]

I have understood, I was wrong. Thank you very much.

[rnsandeep]

Same Issue with me. I changed images to 224 instead of 448.
Resolved it by changing this line in bcnn_finetuning.py
218. self.conv5_3 = tf.reshape(self.conv5_3,[-1,512,784]) ''' Reshape conv5_3 from [batch_size, number_of_filters, height*width]
to self.conv5_3 = tf.reshape(self.conv5_3,[-1,512,196]) # 784 depends on the size of image i guess. so for 224, 196 is correct number.
I think instead of hard coding 784 better to get it from tensors.

[Hedlen]

@rnsandeep ,Thank you for your replay. I will follow your advice. Thanks again!

[YanShuo1992]

@rnsandeep @Hedlen Hi, I also meet the same issue and your answers solve it. However, I still have two questions:

1. Do you I need to change the line 232: self.phi_I = tf.divide(self.phi_I,784.0) ?
2. How to calculate the correct value if I use the images in other size?

Cheers

[rnsandeep]

@YanShuo1992 I can tell for multiples of 224. if you make your image 112 i.e. 224/2 it should be 49( 196/4). if you make it 448 (224x2) it should be 196*4( 784). Not sure about every size.

[YanShuo1992]

@rnsandeep my image is 720720. I find you can output the shape of the conv5_3. It is (?, 512,14,14) when the image size is 224. You can simply store the shape of the conv5_3 as s and replace the 196 with ss.

[jiayugedede]

In bcnn_finetuning.py--> for i in range(total_val_batch): <-- have occured error.It's undefine name.