Skip to content

Latest commit

 

History

History
22 lines (18 loc) · 5.03 KB

LITE_MODEL_ZOO.md

File metadata and controls

22 lines (18 loc) · 5.03 KB
Raw

Models for Mobile and Edge devices

This section contains an exhaustive list of models for mobile and edge devices trained on the CPPE-5 (Medical Personal Protective Equipment) dataset that are currently available and you could start using any of these models very easily.

Notes:

  • All the models can also be trained from scratch using the configs provided in the configs or baselines directory.
  • The metrics are the same as the COCO object detection metrics, more information about this could be found in the original paper.
  • All the model here use the channels first format.
  • We currently do not provide FPS benchmarks for the lite models, we welcome contributions to do so.
Method APbox TF Lite TF Lite (FP 16 quantized) TFJS TFJS (FP 16 quantized) TFJS (uint16 quantized) TFJS (uint8 quantized)
SSD 29.5 bucket bucket bucket bucket bucket bucket
YOLO 38.5 bucket bucket bucket bucket bucket bucket
Faster RCNN 44.0 bucket bucket bucket bucket bucket bucket
FCOS 44.4 bucket bucket bucket bucket bucket bucket
FSAF 49.2 bucket bucket bucket bucket bucket bucket
RegNet 51.3 bucket bucket bucket bucket bucket bucket
TridentNet 52.9 bucket bucket bucket bucket bucket bucket