CV-Assignment

This repository contains the code and resources for a computer vision assignment involving object stability prediction using deep learning models. The project uses various backbones, training scripts, and evaluation methods to achieve optimal model performance.

Project Structure

• backbones/
  Contains the backbone architectures used for model training, such as ResNet and InceptionV4.

• data/
  Stores the training and test data:

  • train/: Directory containing training images.
  • test/: Directory containing test images.
  • train.csv: Training labels and metadata for the training set.
  • test.csv: Test set information.
  • sample-solution.csv: A sample submission file for the test data.

• models/
  Contains experiments from 1-33 including various tuning and adaption techniques, maintaining checkpoints generated during training.

• notebooks/
  Jupyter notebooks used for model evaluation and testing:

  • model_evaluation.ipynb: Notebook for evaluating the performance of trained models and result analysis.
  • test.ipynb: Notebook for testing the models and generate prediction csv files.

• predict_results/
  Stores the prediction results in csv form generated by the trained models.

  1. predict_results_resnet18_singleTask_baseline_kaggle_62.40%.csv is SingleTask ResNet18 prediction results as baseline.

  2. predict_results_exp28_fold7_76.17%_epoch26(kaggle second best).csv is the final submitted version of prediction achieving accuracy of 78.44%.

  3. predict_results_exp30_fold17_79.43%_epoch25(kaggle best).csv is the highest submission achieveing accuracy of 78.65%.

• scripts/
  Shell scripts for running various processes:

  • baseline.sh: Script for running the baseline model.
  • k_fold.sh: Script for training using k-fold cross-validation.
  • run.sh: General script for running training and evaluation processes.

• utils/
  Utility functions and modules used throughout the project:

  • average.py: Helper class for avergaing purpose.
  • options.py: Configuration and command-line argument options.

• baseline_train.py
  Script for training the baseline model.

• train.py
  Main training script used for standard model training.

• train_k_fold.py
  Script for training models using k-fold cross-validation.

• baseline_dataset.py
  Dataset loader for the baseline model.

• dataset.py
  General dataset loader used in the project.

Setup

1. Clone this repository:

   git clone https://github.com/LilyWu06/CV-Assignment-.git
   

2. Download raw data to data/

3. Install dependencies:

   pip install -r requirements.txt
   

4. Enjoy Model Training! : )

Training for recreate baselines

1. resnet18_SingleTask:

   1.1 Change SingleTask boolean in resnet18_SE.py to True
   1.2 Change SingleTask boolean in baseline_dataset.py to True
   1.3 Change SingleTask boolean in baseline_train.py to True
   1.4 Change NETWORK parameter to "resnet18_SE" in baseline.sh
   1.5 Run in Linux/Unix shell: chmod +x scripts/baseline.sh
   1.6 Run in Linux/Unix shell: scripts/baseline.sh
   1.7 Waiting until midnight : )

2. resnet18_MultiTask:

   2.1 Change SingleTask boolean in resnet18_SE.py to False
   2.2 Change SingleTask boolean in baseline_dataset.py to False
   2.3 Change SingleTask boolean in baseline_train.py to False
   2.4 Change NETWORK parameter to "resnet18_SE" in baseline.sh
   2.5 Run in Linux/Unix shell: chmod +x scripts/baseline.sh
   2.6 Run in Linux/Unix shell: scripts/baseline.sh
   2.7 Waiting until midnight : )

3. inceptionv4_SingleTask:

   3.1 Change SingleTask boolean in pretrained_inceptionv4.py to True
   3.2 Change SingleTask boolean in baseline_dataset.py to True
   3.3 Change SingleTask boolean in baseline_train.py to True
   3.4 Change NETWORK parameter to "pretrained_inceptionv4" in baseline.sh
   3.5 Run in Linux/Unix shell: chmod +x scripts/baseline.sh
   3.6 Run in Linux/Unix shell: scripts/baseline.sh
   3.7 Waiting until midnight : )

4. inceptionv4_MultiTask:

   4.1 Use dataset.py and train.py now : )
   4.2 Change NETWORK parameter to "pretrained_inceptionv4" in run.sh
   4.3 Run in Linux/Unix shell: chmod +x scripts/run.sh
   4.4 Run in Linux/Unix shell: scripts/run.sh
   4.5 Waiting until midnight : )