HRNet-MaskRCNN-Benchmark icon indicating copy to clipboard operation
HRNet-MaskRCNN-Benchmark copied to clipboard
verified publisher icon HRNet

miss "NUM_OUTS" in NECK config

Open zimenglan-sysu-512 opened this issue 6 years ago • 5 comments

zimenglan-sysu-512 avatar May 30 '19 08:05 zimenglan-sysu-512

btw, the training speed of configs/hrnet/mask_rcnn_hrnet_w18_1x.yaml is too slow, compared with /configs/e2e_mask_rcnn_R_50_FPN_1x.yaml using single gpu with 2 images per iteration (time: 0.5182 (0.4975) v.s. time: 0.3253 (0.3393)).

zimenglan-sysu-512 avatar May 30 '19 08:05 zimenglan-sysu-512

same problem！

haochange avatar Jun 27 '19 05:06 haochange

NUM_OUTS has been fixed

wondervictor avatar Jul 13 '19 07:07 wondervictor

@zimenglan-sysu-512 @haochange HRNets have more convolutions than ResNets, which cause more time consumption.

wondervictor avatar Jul 13 '19 07:07 wondervictor

@zimenglan-sysu-512 @haochange HRNets: The runtime computation cost in Pytorch is higher because multi-branch convolutions are run sequentially. Instead, in tensorflow, the runtime cost is much lower.

welleast avatar Jul 13 '19 09:07 welleast