asesorov / CompAR

Diploma project: retail product detection, recognition and recommendation (Android)

This is my HSE diploma work: Supermarket products detection, recognition and recommendation app for Android.

Dependencies:

• https://github.com/Tencent/ncnn
• https://github.com/nihui/opencv-mobile

Models:

• YOLOv8, trained on SKU110K dataset (75% acc)
• MobileNet-v3 (with lion optimizer), trained on RP2K dataset (93% acc)
• ProductHash, my own model based on pre-trained MobileNet-v3 features (hash size: 256) and my own Lenta dataset (~100% acc)

Screenshots:

References:

• https://github.com/FeiGeChuanShu/ncnn-android-yolov8
• https://github.com/nihui/ncnn-android-nanodet
• https://github.com/Tencent/ncnn
• https://github.com/ultralytics/ultralytics
• https://github.com/lucidrains/lion-pytorch
• https://www.pinlandata.com/rp2k_dataset/
• https://arxiv.org/pdf/2006.12634v7.pdf