A question in IG-MSA - Githubissues

caiyuanhao1998 / Retinexformer

"Retinexformer: One-stage Retinex-based Transformer for Low-light Image Enhancement" (ICCV 2023) & (NTIRE 2024 Challenge)

https://arxiv.org/abs/2303.06705

MIT License

828 stars 64 forks source link

A question in IG-MSA #73

Closed Maydaybye closed 4 months ago

Maydaybye commented 4 months ago

out_c = self.proj(x).view(b, h, w, c) Why use view(b, h, w, c)? I tried it with a single image, and this operation reduces the resolution of the input image and duplicates it several times.

caiyuanhao1998 commented 4 months ago

What do you mean by " tried it with a single image"

What did you try? A projection layer, an MSA, or a Retinexformer?

caiyuanhao1998 commented 4 months ago

if you find our repo useful, please help us star it