mobilenet-v1-ssd300.tensorflow (8bit quantized and fine-tuned)

Application:  Single-stage Object Detection

Base model: MobileNet-v1

Framework: tensorflow1.1

Training Information: weights were fine-tuned using TF fake quantization nodes

Quality: The COCO mAP(IoU=0.50:0.95) on 5000 validation images is 23.0%

Precision: 8-bit precision

Is Quantized: Yes,  using fake quantization - i.e., weights appear in float32 but have only 256 unique values. 

Dataset: COCO val-2017