ShelfWise results, v1

[rafalcycon]

Hey,

Thank you for hosting a leaderboard on this topic.

As per the Leaderboard section in README we (ShelfWise.ai) would like to submit the results of our model:

Dataset	Method	AP	AP.50	AP.75	AR300	AR.50300
SKU110K-Test	ShelfWise v1	0.529	0.892	0.583	0.598	0.926

The CSV file with our model's predictions is available here: https://drive.google.com/file/d/1rOr4E9koXGyEoLqGuRMuemmBMbbV2KdP/view

The method is an SSD with FPN (which also makes it significantly faster than Faster-RCNN), employed with a few mechanics that were developed to specifically tackle dense scenes that are common in retail. As of now we'd like to not disclose our methodology in full detail as it's still work-in-progress and remains in active research.

[skrish13]

Thanks for contributing! :) I've added it. Good to know that SSD works well. Just curious, was this trained on the SKU110K dataset?

PS for others: I double checked the results by using the csv and its correct :)

[rafalcycon]

Thank you! Glad you also added Eran's results from their improved model. Btw. I would be grateful if you could change the reference (currently it's [3]) in our method's name - either to this issue or to our website www.shelfwise.ai ,due to lack of a better reference at this moment.

Yes, it was trained only on the SKU110K dataset.

[skrish13]

Oh it was supposed to be [4], my bad. Have changed it now (to this issue).

[rafalcycon]

Hey Sri,

We discovered a bug in our code 🤦‍♂️ . After fixing it and retraining the model, the following results are returned:

Dataset	Method	AP	AP.50	AP.75	AR300	AR.50300
SKU110K-Test	ShelfWise v1	0.544	0.904	0.606	0.614	0.934

CSV: https://drive.google.com/file/d/1berCzA2cZ5SGmKuPohr73GUyBymlCLhC/view?usp=sharing

Apologies for the mix-up and for bothering you. We'd be grateful if you could update the results if you have a spare minute.

[skrish13]

No issues :) I have updated it now.

[rafalcycon]

Thank you!