https://github.com/NoListen/mxnet-reinspect

https://github.com/NoListen/mxnet-reinspect

Last synced: 2 months ago
JSON representation

• Host: GitHub
• URL: https://github.com/NoListen/mxnet-reinspect
• Owner: NoListen
• License: gpl-3.0
• Created: 2016-08-16T12:00:35.000Z (over 8 years ago)
• Default Branch: master
• Last Pushed: 2018-01-24T19:45:54.000Z (almost 7 years ago)
• Last Synced: 2024-08-01T22:41:17.097Z (5 months ago)
• Language: Python
• Size: 26.9 MB
• Stars: 5
• Watchers: 2
• Forks: 0
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• Awesome-MXNet - Reinspect

README

        
# mxnet-reinspect

MXNet version of [Reinspect](https://github.com/Russell91/reinspect)

# Features

- No training included

- 35 fps for detection (Nvidia TITANX)

- enlarge the bbox to the whole body by scales varying from scenes to scenes

- ROI Pooling for bounding boxes to extract features

# Additional Requirements

- Transfer caffemodel to MXNet using caffe_converter in mxnet/tools.

# Illustration

- mxnet_track.py is the main file

- config.json describes the hyperparameters

- reinspect.json is the network architecture file

- utils is similar to that in [Reinspect](https://github.com/Russell91/reinspect) with redundant files moved.

- model-transfer specifies the needs to deal with the model learned in [Reinspect](https://github.com/Russell91/reinspect)

# Process

- mxnet model load googlenet params

- mxnet model load lstm params (lstm.h5)

- output the proposals

- extract the features of bbox proposals

# Note

- tracking is not included which you can refer to [MOT_XCODE](https://github.com/HansSJTU/MOT_XCODE).

- bridging python and C++ is not included which you can refer to [numpy-opencv-converter](https://github.com/spillai/numpy-opencv-converter).

# TODO

- add learning process to mxnet-reinspect.