Closed liber145 closed 8 years ago
This issue comes from the inconsistent data number in data_layer.cpp. It is stated in the more detailed Readme file of DeepDrivingCode_v2.zip at http://deepdriving.cs.princeton.edu/
Notice: when using your own dataset to train a new model, the data_layer.train needs modification, at least line 77: "key=random(484815)+1; // MUST be changed according to the size of the training set"
Do you know how to edit the value of "key=random(484815)+1; // MUST be changed according to the size of the training set",I repalce the “TORCS_Training_1F” in floder "pre_trained" with the torcs_trainset.zip, (63GB),then I0607 22:34:14.984568 24099 net.cpp:113] Setting up data Segmentation fault (core dumped)
I met segment fault problem when I train deep driving model on collected data. Could you help me?
I have followed Readme to make TORCS and Caffe. Since the training data set you provided is too large for me, I would like to collect a small number of images and train a model on it. The following is what I do:
Collect data:
chenyi-
prefix road.chenyi-
prefix AI traffic cars.torcs_data_collector.sh
and select window & typep
.Esc
to quit above window. TypeEsc
to quit the game. This generatesTORCS_Training_1F
folder.Train TORCS:
data_layer.cpp
todata_layer.run
and changedata_layer.train
todata_layer.cpp
.TORCS_Training_1F
folder inpre_trained
with currentTORCS_Training_1F
../torcs_train.sh
.Then I met segment fault: