atalrefai / car_classifier

Comprehensive car classifier using TensorFlow

Car Classifier

Car classifier project is inspired from the Grab "AI FOR S.E.A." competition

*Note: Please read the README.pdf for more detailed documentation and explanation.

Introduction:

Here, car make and models can be classified immediate with any input images that belong to the provided car class dataset family which are 196 in total. This project uses a lightweight model that allowed to produce faster result and accurate result at the same time.
Thank you for giving such opportunity and let’s get ready to classify!

Dependencies:

You’ll need have the following installed:
• Python3
• Opencv
• Tensorflow v1.13 (latest-recommended)
• Pillow
• Numpy
• CSV
• Windows/Ubuntu

Trained Model:

Model can be found at: “car_classifier/trained_model/grab_model/”

Full complete model with checkpoint can be downloaded at : https://drive.google.com/open?id=1XpZlvwdRf9vINymxMu4dW4_B8YkZQrVk

How to?

Head into working directory: “car_classifier/”:
Then, this project provides 3 different methods of classification based on your input preferences:

Method 1: Classify and view (images):

This method allows you to view the output generated from the model through a simple GUI:

1. Insert your test dataset (images) into the folder “test_images”
2. Run “classify_image.py” using python
3. To navigate your test photos using (keyboard):
4. Press D – Next Photos
5. Press A – Previous Photos
6. Observe the class prediction and its confidence level in GUI and command log.

*Note: sometimes you need to hold it or press multiple times to navigate

Method 2: Classify into CSV (images): (Recommended)

This method allows you to generate all the prediction output based on your test image dataset into .csv format which include the image file name and 3 top class predictions and its confidence level % as well.

1. Insert your test dataset (images) into the folder “test_images”
2. Run “classify_image_into_csv.py” using python
3. Observe the command log and wait for it to generate all outputs
4. Once successful, proceed to view the “car_classify_output.csv”

Method 3: Car detection (videos):

This method enables its utility feature to predict output in real-time video playback:

1. Insert your test dataset (videos) into the folder “test_videos”
2. Run “classify_video.py” using python with the command:
3. Python classify_video.py --input <video’s file name>
4. Example: “python classify_video.py --input video_1.mp4”
5. *Note: not full path, just the video file’s name.
6. Sit back and observe the output as below:

Conclusion:

Thank you for taking your time in evaluating my trained model with the methods as above. I would like to confess that my model might not be very high in accuracy since I have limited resources in training process. However, I will be improving the model from current time being and please let me know if you have any doubts or questions regarding to my implementation and model.