at first, i'm not from that subject field so I dont know what to do. I'm an ordinary engeneer with basics and unorthodox thinking. I used Claude to help me implement. Its a process of several months and bazills of millions of tokens of work. I can't show or disclose, begging for your understanding.
I can't beat WEBP lossless on kodak test suite, YET; But I'm like 5% worser on RGBa. CMYK's similar. Here, the advantage is me being 5-15x faster as WEBP / other lossless codecs. Not yet fully confirmed, but promissing results WEBP will be knocked too. My lossy compression is better than JPEG sometimes, but always ~10++ times faster. Very big building field right now.
The figures are very promissing on lossless video. On cpu cycles beating every SOTA methods by being upto 15x faster or so. Possible path. My lossy shows better sharpness and details in (very) fast moving scenes (1080p, 2k, 4k incl. hdr), while being only ~5-10% bigger than heif and avc, and others with comperable compression params. May be is resolved in x-xx weeks.
→ noticed, my lossless works better as SOTA (JPEG LS, JPEG 2000 and JPEG XL, each lossless mode) on medical imaging / DICOM, voluminetric medical/scientific data, handles pixel bitdepths >32bit; >32000x32000px (jpeg didnt even consume). It seems to scale res/depth without issues. Runs on data from various MRT/CT Imagers and formats found for download show ~15%-50% percent better compression as SOTA/DICOM, at a fraction of cycles. Some results:
CT volumes, ratio and (encode / decode seconds)
- pancreas, 185 slices → 3.26 (0.7 / 0.6) │ JPEG-LS: 2.69 (2.3 / 1.8) │ JPEG 2000: 2.71 (10.6 / 8.5) │ FFV1: 2.70
- COVID chest, 2464 slices → 4.48 (10.0 / 8.6) │ JPEG-LS: 3.30 (24.6 / 18.9) │ JPEG 2000: 3.12 (129 / 101) │ FFV1: 3.21
- Colonography, 950 slices → 2.79, was 2.59 │ JPEG-LS: 2.73 (10.9 / 8.8) │ JPEG 2000: 2.61 (55.6 / 44.2) │ FFV1: 2.99
- LIDC lung, 764 slices, 12-bit → 2.67 (3.1 / 2.8) │ JPEG-LS: 2.40 (9.9 / 7.7) │ JPEG 2000: 2.40 (46.1 / 37.9) │ FFV1: 2.37
- TCGA kidney, 820 slices → 3.31 (3.3 / 2.9) │ JPEG-LS: 2.87 (9.2 / 7.1) │ JPEG 2000: 2.74 (48.1 / 37.3) │ FFV1: 2.80
MR and SAR data is also a possible play field with some prommissing results I've already seen.
So, as everyone too, i want $$$. I decided to walk my own way after being fired in 2 mins two years ago and I dont have much financial reserves left, without getting income atm. Can't start employment; Need & Want to fight with Claude 23h/7d instead on other ideas I'm working on since years. Need MONEY!! So, if you would be me. Open Sourcing the algorithm is bread-less and doesn't pay my/your bills.
I'm seeking Advice or Ideas on how would you approach it, if you were me - with max profit in the head (.. but still do good!).
The actual and future storage/memory crisis opens up a lever. My approach puts together: storage AND compute can both be siginificantly reduced.
- AWS health services, google health and some other smaller cloud service providers
- Medical imaging device producers (mrt/ct) saving on cpu and memory
- Physicians using SaaS for storage/viewing
- Astro/Scientific imaging and processing
- Possible space/SAR/satelite imaging/data, not giving up on it yet.
But how to approach?? If you give it to x, then y will need it too for processing x's data. This gives a conflict in interest. Interest is profit max. What could end up as a strategy? How to find someone to pitch? Selling? Licensing? Other models? Patenting may not work / doesn't use patented methods.
Appreciate any answer and 'thank you' to everyone reading. If I dont respond to your comment, think of me saying 'thank you' while bowing very deeply, weaving my hat. This post is not meant to let others do my work, its a scream for direction - kind of desperate & blind