a.k.a German Traffic Sign Recognition Benchmark :de: :no_entry: :no_bicycles: :no_entry_sign: ...
Use Torch to train and evaluate a 2-stage convolutional neural network able to classify German traffic sign images (43 classes):
This repository has been forked from the Moodstocks/gtsrb repository. It contains the code used to approach the GTSRB challenge as suggest in "Goal" above.
To execute this code,
> ./download.sh
> th run-cnnDropOut2.lua -save results-cnn-do2 -optimization SGD -learningRate 1e-3 -learningRateDecay 1e-7 -momentum 0.9 -plot
To test the results after a given iteration (here iteration 5), use the script "scripts/eval.lua" as follow:
> th scripts/eval.lua results-cnn-do2/hyp_epoch5.csv data/GT-final_test.csv
Traffic Sign Recognition with Multi-Scale Convolutional Networks, by Yann LeCun et al.
http://benchmark.ini.rub.de/Dataset/GTSRB_Final_Training_Images.zip
(263 MB)
http://benchmark.ini.rub.de/Dataset/GTSRB_Final_Test_Images.zip
(84 MB)
http://benchmark.ini.rub.de/Dataset/GTSRB_Final_Test_GT.zip
(98 kB)