How to run

./run.sh

How to set default gpu?

modify run.sh

Test policy?

• see Here, now using gtFine_trainvaltest validation set.
• user name [email protected]
• pw 16720

Train and test files?

• Train files loaded in main.py, using 50000 synthetic images and 15 channel of ground truth (1 segmentation + 14 joints)
• Test files loaded in test_cityscapes.py, using 500 real images from CityScape validation set.

What's inside logs?

• CityScape IoU testing result of every epoch

Useless files

• test_mp4.py (no longer using a test video)
• test_image/* (no longer using several images to test)

Feb 28th

Add

1. Asynchroize data loading (reducing half of the training time, using TFRecord)
2. All-new CPM model and pretrain (Using some combination of old CPM and new CPM)

new TFRecord training data link

scp -r /mnt/sdc1/shangxuan/synthetic3_tf/*

new Pretrain model download link

https://github.com/ppwwyyxx/tensorpack/tree/master/examples/ConvolutionalPoseMachines

Feb 21th

Add

1. load original CPM pretrain parameter
2. Debugged formal evaluation of CITYSCAPE
3. Changed the network structure from CPM to Multi-person CPM (see Here)

Feb 12th

Add

1. Use new training dataset from Unity
2. Formal evaluation from CITYSCAPE
3. New file structures and class definitions
4. Save/load pre-trained models

Jan 28th

Add

1. global pooling layer -> from ParseNet

Corrected

1. testing process, wrong output filename & results
2. change 'SET_VISIBLE_GPU' to explicit constraint