miss "NUM_OUTS" in NECK config

HRNet / HRNet-MaskRCNN-Benchmark

Object detection with multi-level representations generated from deep high-resolution representation learning (HRNetV2h).

MIT License

138 stars 37 forks source link

miss "NUM_OUTS" in NECK config #1

Open zimenglan-sysu-512 opened 5 years ago

zimenglan-sysu-512 commented 5 years ago

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)).

haochange commented 5 years ago

same problem！

wondervictor commented 5 years ago

NUM_OUTS has been fixed

wondervictor commented 5 years ago

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

welleast commented 5 years ago

@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.