Dataset Quality Control

Link to this sectionThe seven-check QC pass#

Run all seven before any training run. None of them take more than 10 minutes — the preprocessing annotated data guide covers many of these in detail:

Link to this sectionSpot-check by sampling#

You don't need to review every image. A 100-sample stratified audit catches systemic issues:

• 10 from each class, picked at random.
• 10 background images.
• 10 "edge cases" from your spec.
• The lowest 10 confidence predictions from a quick pretrained-model run on the train set.

The last bucket is gold: it surfaces the images the model finds unusual — usually mislabeled or genuinely odd.

Link to this sectionClass balance: what to do when it's broken#

Real datasets are imbalanced. The question is whether it'll hurt training.

   forklift  ████████████████████████ 8000 instances
   person    ████████████████ 5400
   pallet    ██ 700                              ← red flag

A 10:1 imbalance like the one above will produce a forklift-confident model that misses pallets. Three fixes, in order of preference:

1. Collect more pallets. Always the best answer. Aim for ≥ 1500 images and ≥ 10,000 labeled instances of every class, not just the easy ones.
2. Oversample at training time. Augment minority-class images more aggressively so each epoch sees them more often.
3. Class-weighted loss (advanced). Don't reach for this until 1 and 2 are exhausted.

Link to this sectionDedup once more, after labeling#

Even after upload-time dedup, post-labeling dedup catches a different kind of duplicate: visually distinct frames with identical labels. They're usually:

• Two camera angles of the same event with the same labels.
• Sequential video frames where nothing moved.

Drop them — they pad volume without adding training signal.

Link to this sectionThe leakage check is not optional#

Random per-image splits leak whenever images are correlated — adjacent video frames, burst captures, or near-duplicate shots of the same moment. Lesson 6 covers the splitting strategy in depth — but the QC version of the question is:

Do adjacent or near-duplicate frames from the same scene/clip end up in both my train and val splits?

If yes, your val mAP is overstating reality, and the model that ships will be worse than the one you measured. The fix is a grouped split — train and val never share a scene/clip key. (Reusing the same camera across splits is fine — the problem is near-identical frames, not the camera itself.) The preprocessing annotated data guide walks through the mechanics; the splits lesson next covers the strategy.

Link to this sectionWhy fixing labels beats tuning hyperparameters#

A common pattern: model val mAP is 0.62, team spends two weeks on hyperparameter tuning, mAP gets to 0.64. Then someone audits the data and finds 8% of forklift labels are mis-classified as pallet. They fix the labels and retrain — mAP jumps to 0.78.

This is the rule of thumb: before changing anything else, fix the data. The fine-tuning guide, the YOLO performance metrics guide, and the preprocessing annotated data guide all reinforce it. Models are good at generalizing from clean signal; they can't generalize from noise.

Link to this sectionNote what you checked#

Before training, jot down what you checked and fixed — a few lines is plenty:

Dataset: forklift_v1
Audit date: 2026-05-09
Total images: 4,820   |   Total labels: 18,442
Background images: 380 (7.9%)
Per-class instances:
  forklift  6,801
  person    8,310
  pallet    3,331
Visual audit: 120 images sampled, 4 mislabeled (3.3%), fixed.
Duplicates removed: 312
Leakage check: split-by-day, no scene overlap.
Class imbalance: pallet ~2.5× lower than person; ok, not collecting more this round.
Ready to train: yes.

These notes are also handy if a customer or auditor later asks what you checked — "yes, here's what we looked at" beats "yes, we looked at it" with nothing written down.