Ecosyste.ms: Timeline

Browse the timeline of events for every public repo on GitHub. Data updated hourly from GH Archive.

YYY7C

YYY7C starred HapKoM/coco-faces
YYY7C starred VisionMillionDataStudio/BlackBeard412
YYY7C starred ethan-li-coding/SemiGlobalMatching
YYY7C starred ultralytics/ultralytics
YYY7C opened an issue on Tencent/ncnn
layer ReduceL2 not exists or registered
## error log | 日志或报错信息 | ログ 转换模型后,L2标准化层运行报错:layer ReduceL2 not exists or registered ## model | 模型 | モデル 1. 报错部分在onnx位置: ![image](https://github.com/user-attachments/assets/d84dd775-1112-4250-a...
YYY7C closed an issue on Fediory/HVI-CIDNet
有关HSL和HVI
非常棒的工作,在看完论文后对HVI的效果很感兴趣,个人认为HVI和HSL有点相同,但是HVI需要额外的训练,如果使用HSL进行代替是否可行?
YYY7C created a comment on an issue on Fediory/HVI-CIDNet
> Q1:HVI是否需要额外的训练? A1:这个其实不需要,因为他是和CIDNet同时训练的。我们只是增加了很少的参数,这个参数和后面的网络结构训练是合并起来的,不需要先训练HVI,再训练CIDNet。当然,多了这一个参数只会增加一点点时间和空间成本,这对于训练网络结构来说可以忽略不计。如果我们把这些HVI的参数都确定下来,那么其实和只训练CIDNet就没区别了,相当于是把图像换了一个颜色空...

View on GitHub

YYY7C opened an issue on Fediory/HVI-CIDNet
有关HSL和HVI
非常棒的工作,在看完论文后对HVI的效果很感兴趣,个人认为HVI和HSL有点相同,但是HVI需要额外的训练,如果使用HSL进行代替是否可行?
YYY7C starred wangchx67/SYSU-FVL-T2
YYY7C created a comment on an issue on anliyuan/Ultralight-Digital-Human
> > > 我试过加上optimizer.zero_grad(set_to_none=True)后,loss会很小,偶尔会报错。尝试改动train.py中相应的10*sync_loss,区别不太大。也想知道作者大大累加梯度的本意 > > > > > > 加上optimizer.zero_grad()后训练大约十来个epoch就可以收敛了,因为只有单个样本。 同时我注意到在整个网络的...

View on GitHub

YYY7C created a comment on an issue on anliyuan/Ultralight-Digital-Human
> 我试过加上optimizer.zero_grad(set_to_none=True)后,loss会很小,偶尔会报错。尝试改动train.py中相应的10*sync_loss,区别不太大。也想知道作者大大累加梯度的本意 加上optimizer.zero_grad()后训练大约十来个epoch就可以收敛了,因为只有单个样本。 同时我注意到在整个网络的训练中也进行了梯度累积,尝试清零后...

View on GitHub

YYY7C opened an issue on anliyuan/Ultralight-Digital-Human
About SyncNet
为什么 SyncNet训练时不用进行optimizer.zero_grad()?
YYY7C starred KarelZhang/RUAS
YYY7C starred advimman/lama
YYY7C starred LZ-CH/Exposure_Correction-pytorch
YYY7C closed an issue on CyberAgentAILab/TANGO
Some questions regarding graph construction and feature matching
Great work! After reading the paper, I'd like to ask a few questions: 1.The construction of nodes comes from four video frames. Does it mean that a node is composed of these four frames of images?...
YYY7C starred eoffermann/HDTF
YYY7C starred Fediory/HVI-CIDNet
YYY7C starred rockchip-linux/rknpu
YYY7C starred amtoaer/bili-sync
YYY7C starred YudongGuo/AD-NeRF
YYY7C opened an issue on CyberAgentAILab/TANGO
Some questions regarding graph construction and feature matching
Great work! After reading the paper, I'd like to ask a few questions: 1.The construction of nodes comes from four video frames. Does it mean that a node is composed of these four frames of images?...
YYY7C starred vis-opt-group/SCI
YYY7C starred AaronComo/LipFD
YYY7C starred Captain1986/CaptainBlackboard
YYY7C starred anliyuan/Ultralight-Digital-Human
YYY7C starred CVI-SZU/UniFace
YYY7C starred elvisyjlin/AttGAN-PyTorch
YYY7C starred Sanoojan/REFace
YYY7C starred HaiyuWu/SOTA-Face-Recognition-Train-and-Test
Load more