lan666as / scl-jogjas-island

Our team solution for Shopee Code League competition.

SCL-Jogjas-Island

Our team solution for Shopee Code League competition.

Week 1 - Order Brushing

Solution

We use sliding window concept and some data manipulation.

Result

Our solution manages to get full score (1.0) on the private leaderboard. You can run solution.cpp to get the submission CSV.

Week 2 - Product Detection

In this competition, a multiple image classification model needs to be built. There are ~100k images within 42 different categories, including essential medical tools like masks, protective suits and thermometers, home & living products like air-conditioner and fashion products like T-shirts, rings, etc. For the data security purpose the category names will be desensitized. The evaluation metrics is top-1 accuracy.

Solution

We use three models (EfficinetNet-B4, EfficientNet-B5, and Xception) with no data augmentation and unfreeze last few layers for each model.

Result

This solution manages to get 0.83922 accuracy in the private leaderboard (11th place out of 823 teams).

Pre-trained Models

You can get our submission CSV by running Inference.ipynb with our pretrained models in here.