Digital Africa Plantation Counting Challenge
Helping Côte d'Ivoire
$10 000 USD
Challenge completed over 2 years ago
Prediction
Computer Vision
Object Detection
701 joined
219 active
Start
Feb 23, 23
Close
Apr 09, 23
Reveal
Apr 09, 23
More than 800 completely empty images
Data · 20 Mar 2023, 01:40 · 2
Hello,
I have found many completely black images in the dataset:
- 581 in the training set
- 224 in the test set
What is the point of including such images in the dataset?
All those images in the training set have a target of 0 except one with a target of 29!
As a side note, I tried setting to zero the predictions in the test set for those images but my score got worse. This is just another hint to labeling problems in the dataset.
Also based on my experiments there are many miss labeled images in the test set and this will potentially ruin the leaderboard. Can someone on Zindi please address this problem ?
I agree. There are so many mislabeled images.
I also noted other non-empty images have target 0. My assumption on those was that they are not cocoa tress. But checking through, I realized there are images with many cocoa trees, yet the have target given as zero.