Training error

[f-e-l-i-x]

Hi,

Training using the training branch causes the following error:

sudo ./train.sh -d '/####/sessions/2018_01_22__10_31_06_PM' -n "#model" -m categorical
Using TensorFlow backend.
[ 28 9 9 115 40 77 169 2578 1721 1616 1751 1648 1759 1105
380]
[-30. -26.02 -22.04 -18.06 -14.08 -10.1 -6.12 -2.14 1.84 5.82
9.8 13.78 17.76 21.74 25.72 29.7 ]
Input shape: (99, 132, 3)

Layer (type) Output Shape Param #

conv2d_1 (Conv2D) (None, 48, 64, 24) 1824

conv2d_2 (Conv2D) (None, 22, 30, 32) 19232

conv2d_3 (Conv2D) (None, 9, 13, 64) 51264

conv2d_4 (Conv2D) (None, 7, 11, 64) 36928

flatten_1 (Flatten) (None, 4928) 0

dense_1 (Dense) (None, 100) 492900

dropout_1 (Dropout) (None, 100) 0

dense_2 (Dense) (None, 50) 5050

dropout_2 (Dropout) (None, 50) 0

angle_out (Dense) (None, 15) 765

Total params: 607,963
Trainable params: 607,963
Non-trainable params: 0

None
Best model saved in /#/model-model-adam-100-50-Thu Jan 25 14:16:50 2018.h5
2018-01-25 14:16:52.415660: I tensorflow/core/platform/cpu_feature_guard.cc:137] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
Epoch 1/200
Traceback (most recent call last):
File "/home/flx/projects/burro-trainer/burro/burro/train.py", line 38, in
main()
File "/home/flx/projects/burro-trainer/burro/burro/train.py", line 35, in main
train_categorical(data_dir, model_name)
File "/home/flx/projects/burro-trainer/burro/burro/trainers/trainer.py", line 101, in train_categorical
callbacks=[tb, model_cp, e_stop])
File "/usr/local/lib/python2.7/dist-packages/keras/legacy/interfaces.py", line 87, in wrapper
return func(*args, **kwargs)
File "/usr/local/lib/python2.7/dist-packages/keras/models.py", line 1117, in fit_generator
initial_epoch=initial_epoch)
File "/usr/local/lib/python2.7/dist-packages/keras/legacy/interfaces.py", line 87, in wrapper
return func(*args, **kwargs)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1840, in fit_generator
class_weight=class_weight)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1559, in train_on_batch
check_batch_axis=True)
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 1234, in _standardize_user_data
exception_prefix='input')
File "/usr/local/lib/python2.7/dist-packages/keras/engine/training.py", line 140, in _standardize_input_data
str(array.shape))
ValueError: Error when checking input: expected conv2d_1_input to have shape (None, 99, 132, 3) but got array with shape (256, 120, 160, 3)

---

keras version is 2.0.6
tensorflow is 1.4.1

I am pretty clueless here

Felix

[f-e-l-i-x]

I edited /home/####/burro-trainer/burro/burro/trainers/trainer.py

remove: input_shape=(99-config.camera.crop_top - config.camera.crop_bottom, 132, 3)
add: input_shape=(120,160,3)

That seems to work for now

[yconst]

Yea the input shape is hardcoded for now. Needs to read from settings. Keep in mind that this is in two locations in trainer.py: One in line 37 and one in 108 for classification and regression correspondingly