Activity
Stars and mentions · Last 30 days, Oct 6 – Nov 4
Top repositories
richgel999’s repositories ranked by pull request authors, as on trending.
Contributes to
Repositories outside richgel999’s account where richgel999 opened or reviewed pull requests in the past 365 days.
Who noticed richgel999
Everyone who starred or forked richgel999’s repositories since the archive began in 2011.
Notable accounts
Ordered by the stars their own repositories have received.
Biggest fans
Accounts that starred the most of these repositories.
Breakout weeks
Each repository’s best week of new stars, with the release published around it.
Achievements
Awards earned by the repositories richgel999 owns. Select an award to embed its badge.
No awards yet
Repositories earn awards by reaching the top three in a completed trending day or week.
A repository earns an award for finishing first, second or third in a completed trending day or week, ranked by new stars, new forks or pull request authors. Awards follow a repository through renames and transfers and count toward its current owner. Weeks run Monday to Sunday, UTC.
Mentions
Another 4x neural upsampling example. Left: PS2/Xbox era (512x512) Right: 4x upsampled, 12.5k weights (2048x2048) ~250 lines of C++ code. richgel999/neural_upsampler t.co
@jdamdami Back in the real world I'm shipping neural upsamplers and JPEG filters, GPU texture encoders, etc. with AI. richgel999/neural_upsampler
If you want to mess around with neural JPEG artifact reduction, the tiny C++ library I handed off to Codex for integration is here: richgel999/jpegc_deblock "A 7x7-window MLP (52 inputs: the luma window as differences from the centre, the centre's luma, the 8x8 block phase, the quality factor; hidden 32 and 16, leaky ReLU; one linear output) that predicts a luma correction added to R, G and B, equivalent to a Y-only correction in YCbCr. Trained by finite differences on 514 PNGs round-tripped through baseline JPEG at qualities 30..95. On held-out images it gains about +0.95 dB luma PSNR at quality 30, +0.9 at 40, +0.7 at 80. About 20 Mpx/s on 32 threads, 1.6 Mpx/s on one."
I started AI on 24/7 research on this problem months ago: Once you figure out how MLP's work, you'll realize they aren't always necessary. This is neural texture compression - but only in 1 layer and without activation functions: richgel999/nntc 1 layer = only 1 GEMV to decode all samples. This supports Cooperative Vectors etc.
I'm able to deploy Codex on my older codebases and make weeks of progress happen in hours. An MCP server for this dataset is now live. The next step is geocoding and real location-based searches. I'm tackling this smaller dataset before UFOCAT. richgel999/ufo_data