Core Roadmap

[pkunray]

1. Dataset

Yoga Posture Dataset

2. Feature Engineering

Different treatments according to different algorithms

3. Image Classification Algorithm candidates

• KNN
• Decision Tree
• Random Forest
• SVM
• CNN
• Transformer

4. Performance Evaluation & Comparison

• Accuracy
• Confusion Matrix
• Recall

Reference Projects: https://www.kaggle.com/datasets/tr1gg3rtrash/yoga-posture-dataset/code

[pkunray]

@hatamoto429 @hatamoto429 @Alex-exy @stickrundude

I showed this to Max after class, he approved this idea.

Other questions I asked:

Q: Do we need to run the training process during the presentation? A: No, just show the code, a good idea is to save a trained model, and test it with real data

Q: Does using a simple algorithm necessarily result in a low grade? A: No, but the person picking KNN might be asked questions about CNN, the grade depends both on the project and the answers

Q: In addition to indicators such as accuracy score, what else can we compare? A:

• Training time
• Differences in data processing

Q: Other suggestions? A: Show how your model is improved, have good conclusions

[pkunray]

1. Same Dataset

2. Two approaches

   • CNN (Alex, Lukas)
   • Transformer (KshitiJ, Kun)

Deadline: 2024-07-11 2pm