Version: 0.4

Summary

We added detection for 17 new logos and improved detection on 42 others. Here is a complete list of our logos mapped to their brand/product identifiers.

Additions & Enhancements

We added 17 new logos. Here are some of them

  1. ICE
  2. NYSE
  3. Target
  4. Hollister
  5. Abercrombie & Fitch
  6. Paya
  7. UPS

Enhanced logo detection 42 brands. Here are some of them

  1. Bank of America
  2. Chase
  3. Impots
  4. Ebay
  5. Binance
  6. Barclays
  7. Capitolone

Complete list of logos here.

Model details

One common observation while training new models (see data from previous run here),  YOLOv5 adapts to the training dataset faster than its predecessors. We usually get our best model by the 200th iteration. After that, we just see incremental gains.

Our validation loss started increasing after the 350th iteration which is a bad sign. This could mean trouble if you are not storing the best model/ do not have early stopping in place.

Mean Average Precision - Higher the value better the model
Validation Loss - Lower the value better the model
Recall - Higher the value better the model

Sample Results

Below are a few results from the new logo detection model