Zehao Shi

YellowFin optimizer has been intergrated.

2

I have no gpu resources to train on the imagenet. I have validated it on mnist dataset, got 97.64% accuracy. Call for training \~_\~ Best, Zehao

Hi, I'm trying NearPy, and I got some error. Below is some code and error message. ``` import scipy.io as sio mat_file = sio.loadmat("C:\\Users\\RA2\\Desktop\\Workspaces\\LightenedCNN_A_lfw.mat") fea = mat_file["features"] labels = mat_file["labels_original"]...

syncbn with "channel_last=True" produce wrong result when feature_num is not pow-of-two

1

**Describe the Bug** When feature_num is not pow-of-two, apex.parallel.SyncBatchNorm will produce wrong result. I test it step by step, and found it produce wrong mean and var when feature_h and...

In detectron, MAX_ITER: 180000, IMS_PER_BATCH: 1, NUM_GPUS: 8, results in 180000*1*8 = 1440000. While in this repo, nr_image_per_epoch = 80000, max_epoch=30, results in 80000*30=2400000. Reference: https://github.com/facebookresearch/Detectron/blob/master/configs/12_2017_baselines/e2e_faster_rcnn_R-50-C4_1x.yaml