InvalidArgumentError (see above for traceback): assertion failed: [] [Condition x == y did not hold element-wise:] [x (losses/fast_rcnn_cls_loss/SparseSoftmaxCrossEntropyWithLogits/Shape_1:0) = ] [101 1] [y (losses/fast_rcnn_cls_loss/SparseSoftmaxCrossEntropyWithLogits/strided_slice:0) = ] [101]

[longzeyilang]

Hi,
I'm trying to train clutteredMNIST of gray images with ResNet 50 V1 model, using the command "python faster_rcnn_conv5.py -n 10 -e 20 -y 'clutteredMNIST.yml' ",however the error below occured
"InvalidArgumentError (see above for traceback): assertion failed: [] [Condition x == y did not hold element-wise:] [x (losses/fast_rcnn_cls_loss/SparseSoftmaxCrossEntropyWithLogits/Shape_1:0) = ] [101 1] [y (losses/fast_rcnn_cls_loss/SparseSoftmaxCrossEntropyWithLogits/strided_slice:0) = ] [101]"

[longzeyilang]

waiting online

[Linggeng]

have you solved this issue?

[fbiying87]

I have the same problem by using faster_rcnn_conv5.py -n 0 -e 5 as given in the example. Has anyone else solved this issue yet? Thanks.

[zyfccc]

Me 2. Running example on MacOS Mojave and python 3.6.4.

[binhmuc]

I also got this problem. I run in python 3.5.2, tensorflow-gpu 1.8.0, My ubuntu version is 16.04

[Ien001]

find
fast_rcnn_cross_entropy=tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=tf.squeeze(fast_rcnn_cls_score), labels=labels)) in loss_functions.py
change to
fast_rcnn_cross_entropy=tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=tf.squeeze(fast_rcnn_cls_score), labels=tf.argmax(labels, 1)))

Would u plz tell us why it works? it truly works! Thanks!

[Rorschaaaach]

在
loss_functions.py中找到fast_rcnn_cross_entropy = tf.reduce_mean（tf.nn.sparse_softmax_cross_entropy_with_logits（logits = tf.squeeze（fast_rcnn_cls_score），labels = labels））
更改为
fast_rcnn_cross_entropy = tf.reduce_mean（tf.nn.sparse_softmax_cross_entropy_with_logits（logits = tf.squeeze） （fast_rcnn_cls_score），labels = tf.argmax（labels，1）））

你能告诉我们它为什么有效吗？它确实有效！谢谢！

不好意思，刚刚看了下这好像是个错误的方法我等会会删除，你可以看一下labels的输出，可能是因为这方面的原因。

[zrfan]

在
loss_functions.py中找到fast_rcnn_cross_entropy = tf.reduce_mean（tf.nn.sparse_softmax_cross_entropy_with_logits（logits = tf.squeeze（fast_rcnn_cls_score），labels = labels））
更改为
fast_rcnn_cross_entropy = tf.reduce_mean（tf.nn.sparse_softmax_cross_entropy_with_logits（logits = tf.squeeze） （fast_rcnn_cls_score），labels = tf.argmax（labels，1）））

你能告诉我们它为什么有效吗？它确实有效！谢谢！

不好意思，刚刚看了下这好像是个错误的方法我等会会删除，你可以看一下labels的输出，可能是因为这方面的原因。

应该是：
fast_rcnn_cross_entropy = tf.reduce_mean(
tf.nn.sparse_softmax_cross_entropy_with_logits(logits=tf.squeeze(fast_rcnn_cls_score),
labels=tf.squeeze(labels)))
吧，因为sparse_softmax_cross_entropy_with_logits要求标签的维度是[batch_size]大小，这里需要把label的形状reshape一下