Skip to content

4xNomos8kSCHAT

Compare
Choose a tag to compare
@Phhofm Phhofm released this 24 Apr 15:20
· 6 commits to main since this release
c664eb5

Releasing two HAT (S and L) models on the Nomos8k_sfw dataset (they added HAT-S code and models two months ago)


Name: 4xNomos8kSCHAT-L
Author: Philip Hofmann
Release Date: 30.06.2023
License: CC BY 4.0
Network: HAT
Architecture Option: HAT
Scale: 4
Purpose: A photo 4x upscaling model based on musl's Nomos8k_sfw dataset for realistic sr
Iterations: 132,000
batch_size: 4
HR_size: 256
Dataset: Nomos8k_sfw
Number of train images: 6118
OTF Training: Yes
Pretrained_Model_G: HAT-L_SRx4_ImageNet-pretrain.pth

Description: 4x photo upscaler with otf jpg compression and blur, trained on musl's Nomos8k_sfw dataset for realisic sr. This is a big model file, upscaling a single image might take a while and use up ressources.


Name: 4xNomos8kSCHAT-S
Author: Philip Hofmann
Release Date: 30.06.2023
License: CC BY 4.0
Network: HAT
Architecture Option: HAT-S
Scale: 4
Purpose: A photo 4x upscaling model based on musl's Nomos8k_sfw dataset for realistic sr
Iterations: 169,000
batch_size: 4
HR_size: 256
Dataset: Nomos8k_sfw
Number of train images: 6118
OTF Training: Yes
Pretrained_Model_G: HAT-S_SRx4.pth

Description: 4x photo upscaler with otf jpg compression and blur, trained on musl's Nomos8k_sfw dataset for realisic sr. A similiar model as above, but as an HAT-S model instead.


Examples
Imgsli Example 1: https://imgsli.com/MTg5MDYy
Imgsli Example 2: https://imgsli.com/MTg5MDY1
Imgsli Example 3: https://imgsli.com/MTg5MDY2
Imgsli Example 4: https://imgsli.com/MTg5MDY3

Example1
Example2
Example3
Example4