Retail Object Detection

NVIDIA TAO Release 30.2205

The retail object detection model detects one or more items within an image and returns a bounding box around each detected item. These retail items are generally packaged commercial goods with barcodes and ingredient labels on them, as seen at a check-out counter.

As part of this model instance, 2 detectors are provided.

  1. Binary class detector

  2. 100-class detector

The binary class detector, detects general retail items and returns a single category.

The 100 class retail detector on the other hand, detects one more objects belonging to 100 specific classes.

These models are based on EfficientDet-D5. EfficientDet is a one-stage detector with the following architecture components:

  • NvImageNet pretrained EfficientNet-B5 backbone

  • Weighted bi-directional feature pyramid network (BiFPN)

  • Bounding and classification box head

  • A compound scaling method that uniformly scales the resolution, depth, and width for all backbone, feature network, and box/class prediction networks at the same time

The primary use case intended for these models is to detect object on a retail checkout counter.

The datasheet for this model is captured in it’s model card host at NGC_.

© Copyright 2022, NVIDIA.. Last updated on Dec 13, 2022.