🔰 Code Repository, Models and Experiments

• [ ] Codebase is well-organized
• [ ] Model naming is clear and intuitive
• [ ] Experiment logs are accurate and detailed Consider MLFlow, W&B or other similar tools and services.
• [ ] Essential metadata for each model is available Dataset version, train script version, and training parameters. Consider using a DVC tool.

📊 Data Preparation and Analysis

• [ ] Original data visualization scripts/tools are used To ensure accurate interpretation and adequacy of labels.
• [ ] Original data analysis is conducted Evaluating characteristics like class count, sample distribution by class, object size distribution for detection, and pixel distribution in masks, among others.

🗄 Datasets and Integrity

• [ ] Data has been converted to an optimal format Consider HDF5 – one of the most convenient formats. To reduce volume and disk load, it is advisable to store data in 8-bit if acceptable.
• [ ] Split into Train and Test has been executed as separate sets Ideally, Test and Validation should also be distinguished.
• [ ] Data in the databases/sets are randomly shuffled
• [ ] The relationship between the original data and the data in the databases is preserved
• [ ] Metadata is associated with the data E.g. attributes in HDF5 store the version of the data generation script, parameters, etc.
• [ ] Developed a script for visualizing data from the database Thereby ensuring the correctness of data storage in the database.

🧮 Evaluating Models