Phips/NERVE

link the converter Space

add links to Space / app / Colab

document 2x OTF GAN + patch/batch result

add 2x OTF GAN (lq128/batch8 finetune) 2x_NERVE_OTF_gan_fp16.ncnn.bin

add 2x OTF GAN (lq128/batch8 finetune) 2x_NERVE_OTF_gan_fp16.ncnn.param

add 2x OTF GAN (lq128/batch8 finetune) 2x_NERVE_OTF_gan_1x3xHxW_fp32_op20.onnx

add 2x OTF GAN (lq128/batch8 finetune) 2x_NERVE_OTF_gan.safetensors

restore nerve_arch.py (was removed by mistake)

remove stray requirements.txt

remove stray nerve_arch.py

remove stray app.py

restore model card (accidental overwrite) + benchmark update

initial NERVE upscale Space (ZeroGPU)

add honest 2026-09-22 speed benchmark update

fix stale comments

fix stale comments

fix stale comments

add icnr.py (ICNR init helper needed by nerve_arch.py)

document icnr.py dependency in quickstart

restore full README (was clobbered by dir upload)

full-res N-vs-H OST_009_nerve_vs_heart_compare.png

full-res N-vs-H 00003_nerve_vs_heart_compare.png

full-res N-vs-H realphoto_nerve_vs_heart_compare.png

full-res N-vs-H noisy_nerve_vs_heart_compare.png

examples: replace overview montages with full-resolution (no downscale)

link Real-ESRGAN N-vs-H comparisons

add input OST_009.png

add input 00003.png

add NERVE-vs-HEART (Real-ESRGAN test set): 00003_nerve_vs_heart_1to1.png

add NERVE-vs-HEART (Real-ESRGAN test set): 00003_nerve_vs_heart_compare.png

add NERVE-vs-HEART (Real-ESRGAN test set): OST_009_nerve_vs_heart_1to1.png

add NERVE-vs-HEART (Real-ESRGAN test set): OST_009_nerve_vs_heart_compare.png

link NERVE-vs-HEART comparisons

add NERVE-vs-HEART comparison: noisy_nerve_vs_heart_1to1.png

add NERVE-vs-HEART comparison: noisy_nerve_vs_heart_compare.png

add NERVE-vs-HEART comparison: realphoto_nerve_vs_heart_1to1.png

add NERVE-vs-HEART comparison: realphoto_nerve_vs_heart_compare.png

remove stray jpg

document 1:1 crops + inputs

examples: add 1:1 native crops + input images

correct arch line count

reorder sections for reading flow

enrich model-card metadata (library_name, tags, dataset)

fix ncnn layout note (not fixed-size)

rewrite Why section: academic tone, trainer motivation, community credit

add 'Why NERVE?' as the second section

add structured 'help wanted' section

document ncnn usage + what we tested / what we did not

soften remaining CoreML claim

no unverified claims, CC0 data section, embedded examples