Trash Classifier

Setup

cd ~
mkdir .envs
cd .envs
mkdir thrash-classifier
virtualenv --no-site-packages -p python3 ./thrash-classifier
cd #project-directory
source ~/.envs/thrash-classifier/bin/activate
pip3 install -r requirements.txt

Adding Python Modules via Pip

pip3 install #new-python-module
pip3 freeze > requirements.txt
cat requirements.txt
git push

Converting Images to TFRecords

python3 ./cnn/image_to_tfrecord.py

Running SVM Classifiers

python3 . svm_inception_bottleneck

python3 . svm_sift_kmeans

python3 . svm_orb_kmeans

python3 . svm_rgb_sift_kmeans

python3 . svm_rgb_gray_sift_kmeans

Running CNN Classifier

python3 ./cnn/cnn_classifier.py -a #alexnet/zfnet

Dataset

We obtained our dataset of 2527 trash images from garythung/trashnet. The pictures were taken by placing the object on a white posterboard and using sunlight and/or room lighting. They have been resized down to 256 x 256 when converting the images to tfrecords. The devices used to take the photos were Apple iPhone 7 Plus, Apple iPhone 5S, and Apple iPhone SE.

CNN Architecture (AlexNet)

Methods in layers module expect input tensors to have shape [batch_size, image_height, image_width, channels].

Under 'SAME' padding scheme, output is calulated as such:

out_height = ceil(float(in_height) / float(strides[1]))
out_width  = ceil(float(in_width) / float(strides[2]))

Under 'VALID' padding scheme, output is calulated as such:

out_height = ceil(float(in_height - filter_height + 1) / float(strides[1]))
out_width  = ceil(float(in_width - filter_width + 1) / float(strides[2]))

1. Input Layer (garythung dataset)

   Output Shape: [-1, 256, 256, 3]

2. Convolution Layer 1

   Input Shape: [-1, 256, 256, 1]
   Filter Shape: [11, 11]
   Number of Filters: 96
   Strides Shape: [4, 4]
   Output Shape (Same Padding): [-1, 64, 64, 96]
   Activation Function: ReLU

3. Pooling Layer 1

   Input Shape: [-1, 64, 64, 96]
   Filter Shape: [3, 3]
   Strides Shape: [2, 2]
   Output Shape (Valid Padding): [-1, 31, 31, 96]

4. Convolution Layer 2

   Input Shape: [-1, 31, 31, 1]
   Filter Shape: [5, 5]
   Number of Filters: 192
   Strides Shape: [1, 1]
   Output Shape (Same Padding): [-1, 31, 31, 192]
   Activation Function: ReLU

5. Pooling Layer 2

   Input Shape: [-1, 31, 31, 192]
   Filter Shape: [3, 3]
   Strides Shape: [2, 2]
   Output Shape (Valid Padding): [-1, 15, 15, 192]

6. Convolution Layer 3

   Input Shape: [-1, 15, 15, 192]
   Filter Shape: [3, 3]
   Number of Filters: 288
   Strides Shape: [1, 1]
   Output Shape (Same Padding): [-1, 15, 15, 288]
   Activation Function: ReLU

7. Convolution Layer 4

   Input Shape: [-1, 15, 15, 288] Filter Shape: [3, 3]
   Number of Filters: 288
   Strides Shape: [1, 1]
   Output Shape (Same Padding): [-1, 15, 15, 288]
   Activation Function: ReLU

8. Convolution Layer 5

   Input Shape: [-1, 15, 15, 288] Filter Shape: [3, 3]
   Number of Filters: 192
   Strides Shape: [1, 1]
   Output Shape (Same Padding): [-1, 15, 15, 192]
   Activation Function: ReLU

9. Pooling Layer 3

   Input Shape: [-1, 15, 15, 192]
   Filter Shape: [3, 3]
   Strides Shape: [2, 2]
   Output Shape (Valid Padding): [-1, 7, 7, 192]

10. Dense Layer 1

    Input Shape: [-1, 7 * 7 * 192]
    Number of Neurons: 4096
    Output Shape: [-1, 4096]
    Activation Function: ReLU

11. Dropout Layer 1

    Input Shape: [-1, 4096]
    Dropout Rate: 0.2
    Output Shape: [-1, 4096]

12. Dense Layer 2

    Input Shape: [-1, 4096]
    Number of Neurons: 4096
    Output Shape: [-1, 4096]
    Activation Function: ReLU

13. Dropout Layer 2

    Input Shape: [-1, 4096]
    Dropout Rate: 0.2
    Output Shape: [-1, 4096]

14. Logits Layer

    Input Shape: [-1, 4096]
    Number of Neurons: 5
    Output Shape: [-1, 5]

List of Tools & Resources

• https://github.com/garythung/trashnet
• https://adeshpande3.github.io/The-9-Deep-Learning-Papers-You-Need-To-Know-About.html
• https://smist08.wordpress.com/tag/adam-optimizer/
• https://stackoverflow.com/questions/36162180/gradient-descent-vs-adagrad-vs-momentum-in-tensorflow?utm_medium=organic&utm_source=google_rich_qa&utm_campaign=google_rich_qa
• https://www.kaggle.com/grfiv4/plot-a-confusion-matrix
• https://medium.com/@utsumukiMutsuki/using-inception-v3-from-tensorflow-hub-for-transfer-learning-a931ff884526
• https://arxiv.org/pdf/1707.07013.pdf
• https://machinelearningmastery.com/report-classifier-performance-confidence-intervals/
• https://arxiv.org/pdf/1505.05424.pdf
• https://scikit-optimize.github.io/#skopt.gp_minimize
• http://scikit-learn.org/stable/modules/generated/sklearn.svm.SVC.html
• http://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html
• https://code.oursky.com/tensorflow-svm-image-classifications-engine/
• https://github.com/GoogleCloudPlatform/tf-estimator-tutorials/blob/master/07%20-%20%20Image%20Analysis/02.0%20-%20CNN%20with%20CIFAR-10%20tfrecords%20dataset.ipynb
• https://docs.opencv.org/3.3.0/da/df5/tutorial_py_sift_intro.html
• https://docs.opencv.org/3.0-beta/modules/features2d/doc/feature_detection_and_description.html#orb-operator
• https://github.com/Fdevmsy/Image_Classification_with_5_methods