We pushed hard this round but fell just short of the Top 100 leaderboard — by only 3 positions! Here’s why:

1️⃣ We used a small model (convnextv2_tiny) for faster experimentation — try larger CNNs and transformer-based models to improve performance.

2️⃣ We trained on just one fold to speed up iteration during the live stream — using all folds will definitely boost scores in future runs.

Update: Training on all folds gets you comfortably to top 100 (Next experiments , change the model and customize the augmentations goodluck)

Huge thanks to @Brainiac and @Sodiq_Babawale_ for joining the streams and showing love , really means a lot 🙏🏽

📆 Catch me next week on Tuesday, Wednesday, and Thursday at 20: 30hrs to 22:00 hrs EAT on Twitch: https://www.twitch.tv/koleshjr/schedule

🔗 Watch the full stream here: https://youtu.be/MF6P0lx91kc?si=Udk5IxT9B-CbgyBC

Notebook Linked in the Youtube Description