Reference URL: github:https://github.com/naisy/realtime_object_detection
2018.10.16
SSD Object Detection Summary:
Remember to take a cursory look at the notes and start training the model
Error: 1, with branch1.5,tensorflow-gpu==1.8 training model in GT730, video memory 2g, can not run, tensorflow-gpu==1.5 no NoMaxSuppressionv3,
2, using the pre-training model ssd_mobilenet_coco_2018_1_28,tensorflowgpu==1.5 training, the model training on cuda9.0,cudnn7.05, the model obtained in the GT730 2fps,
3, using the pre-training model ssd_mobilenet_coco_2017_11_27,tensorflow-gpu==1.5 training, the model on the GT730 18fps,
4, open two terminals at the same time, 7fps, three terminal 4fps,
5, choose Nms_v1 or NMS_V2, in the test notes, there is a description,
6, part of the model with cudnn7.1 training, part with cudnn7.05 training will report CUDNN related errors.
Before training the model and testing to be aware of, clear steps, know every step why do,
Hardware requirements and software requirements have to be considered in advance.
SSD Object Detection Model training and test summary