Retail vision case study
How a retail vision company increased the efficiency of their models by 12%
Our client is a computer vision startup in the retail space that is backed by some of the prestigious names in the start up world. They have some of the biggest retail chains on the planet in their clientele. With the retail industry set to witness a steady growth and companies investing heavily in autonomous store technology, our client is expected to be a leader in this space.
Our client uses CCTV camera footage to understand consumer behavior and enable a complete unsupervised shopping experience. The performance of their models didn't improve beyond a certain level because of the large number of edge cases and uncertainties in the customer behavior.
When our client approached us, they were expanding into a new geography and the shop layout and the clothing style of the customers were very different to what they had trained their model on and this resulted in a drop in the performance of their computer vision models.
We Have experience dealing with edge case scenarios before and we identified 5 scenarios that affected the performance of the model and started annotating those cases first to be fed into their pipeline for training. We started with the most difficult cases and proceeded to the not so extreme cases and this approach led to a quicker improvement in the performance of the model.
We totally annotated about 100,000+ cases with varying difficulties in a week's time that ended up improving our client's CV model substantially.
Some metrics of the project
No of annotated images - 200,000+
Time taken for the work - 1 Week
Savings compared to annotation platforms - $5000+
Improvements in model performance - 12%
DataClap helped us improve our prediction models in a big way. Right from onboarding to support we were extremely happy with the way things went. This is something that we've not experienced with big data annotation platforms.
-CEO of the startup
With our high quality annotation services, our client was able to improve their model and the predcitions. They were able to successfuly implement their project in the new location and it certainly improved the overall performance of their models with edge case scenarios.