jyihuo/traffic-sign-classification

#Traffic Sign Recognition with PyTorch

Step By Step

• Load the data set (see below for links to the project data set)
• Explore, summarize and visualize the data set
• Design, train and test a model architecture
• Use the model to make predictions on new images
• Analyze the softmax probabilities of the new images
• Summarize the results with a written report

---

###Data Set Summary & Exploration

###1. I used the pandas library to calculate summary statistics of the traffic

• Number of training examples = 34799
• Number of testing examples = 12630
• Image data shape = (32, 32, 3)
• Number of classes = 43

####2. Include an exploratory visualization of the dataset.

Let's see the distribution of class in training-set , validation-set:

It seems that all dataset inclued training, validation, test, have the same distribution. It was great? Let's do something nicer.

3. Show the probabilty for each classes of all dataset

• It seems that targets is unbalance
• It seems that within the middle of the chart, something different between training-set. validation-set and test-set. Should we do something helpful likes resample. (I do many experiment. And it shows that the technology of resample is useless... )

###Design and Test a Model Architecture

####1. Preprocessed & Data Augmentation

As a first step I normalized the image data because

• Training the network more faster
• Traing the network more eaier
• Ignoring the death of activation

As a second step, I decided to convert the images to grayscale because

• It can get the higher score 😄

Here is an example of a traffic sign image before and after grayscaling.

As a last step, I do a list of data augmentation:

• Randomly, choosing 20% data to rotate the images with 30 degree
• Randomly, choosing 10% data to Zoom the images
• Randomly, choosing 10% data to shift the images

####2. My model

I use DenseNet121 which is the popular network architecture.

####3. Training DenseNet

Attention: I retrain DenseNet model from scratch.

To train the model, I used this parameters:

• epoch: 30
• batch_size: 128
• lr: 0.001
• lr_decay_step: 0.1/10 Learning rate narrow 0.1 / 10 epoch

And then, I use Adam to optimizer my loss which is calculated by SoftmaxCrossEntropy.

####4. Accuracy fro Traing, Validation, Testing

My final model results were:

• training set accuracy of 0.9999
• validation set accuracy of 0.9588
• test set accuracy of 0.9482

If a well known architecture was chosen:

• I choose DenseNet model
• It was used to training image in ImageNet Challenge, So I think there is no problem to use DenseNet model for traffic sign classification.

###Test a Model on New Images

####1. Choose five German traffic signs found on the web

Here are five German traffic signs that I found on the web:

The first image might be difficult to classify because the background is so large.

####2. Discuss the model's predictions Here are the results of the prediction:

The model was very state-of-art

####3. Describe how certain the model is when predicting on each of the five new images by looking at the softmax probabilities for each prediction.

(Optional) Visualizing the Neural Network (See Step 4 of the Ipython notebook for more details)

####1. Discuss the visual output of your trained network's feature maps. What characteristics did the neural network use to make classifications?