Sorghum -100 Cultivar Identification - FGVC 9

Introduction

This repository contains source code generated by Luminide. It may be used to train, validate and tune deep learning models for image classification. The following directory structure is assumed:

├── code (source code)
├── input (dataset)
└── output (working directory)

The dataset should have images inside a directory named train_images and a CSV file named train_cultivar_mapping.csv. An example is shown below:

input
├── train_cultivar_mapping.csv
└── train_images
    ├── 2017-06-16__12-24-20-930_0.jpg
    ├── 2017-06-16__12-24-20-930_1.jpg
    ├── 2017-06-16__12-24-20-930_2.jpg

The CSV file is expected to have labels under a column named cultivar as in the example below:

image,cultivar
2017-06-16__12-24-20-930.jpg,PI_257599
2017-06-02__16-48-57-866.jpg,PI_154987
2017-06-12__13-18-07-707.jpg,PI_92270

Quick recipe for using this repo with Luminide

• Accept competition rules.
• Attach a Compute Server that has a GPU (e.g. gcp-t4).
• Configure your Kaggle API token on the Import Data tab.
• On the Import Data tab, choose Kaggle and then enter anlthms/sorghum1 (User Dataset).
• Train a model using the Run Experiment menu.
• Launch inference.sh from the Run Experiment tab to create a submission and use submit.sh to upload it to Kaggle.
• Check the leaderboard to see your score!

Additional features

• Use the Experiment Tracking menu to track experiments.
• To tune the hyperparameters, edit sweep.yaml as desired and launch a sweep from the Run Experiment tab. Tuned values will be copied back to a file called config-tuned.yaml along with visualizations in sweep-results.html.
• To use the tuned hyperparameter values, copy them over to config.yaml before training a model.
• For exploratory analysis, run eda.ipynb.
• To monitor training progress, use the Experiment Visualization menu.
• After an experiment is complete, use the file browser on the IDE interface to access the results on the IDE Server.
• To generate a report on the most recent training session, run report.sh from the Run Experiment tab. Make sure Track Experiment is checked. The results will be copied back to a file called report.html.

NOTE: As configured, the code trains on 20% of the data. To train on the entire dataset, edit full.sh and fast.sh to remove the --subset command line parameter so that the default value of 100 is used.

For more details on usage, see Luminide documentation