Environment
wsitools: 0.26.6 (go1.26.2, windows/amd64)
OS: Windows 11 Pro, build 26200
CPU: AMD Ryzen 7 5700U — 8c/16t
RAM: 28.9 GB
Disk: SSD (C:)
Shell: Windows PowerShell 5.1
Codecs: all OK (avif/htj2k/jpeg/jpeg2000/jpegxl/png/webp)
A scripted sweep exercising every subcommand across the sample containers (SVS, COG-WSI, BIF) — convert across containers, crop, downsample, transcode, per-codec re-encode, composed transforms, region, extract, hash, dump-ifds, and associated-image edits. All 52 operations succeeded (0 failures).
Full per-command results (CSV) and the complete run transcript are attached:
https://gist.github.com/cornish/bb15aa58a899ef23b956ddb4fec855ca
Regression check (all green on Windows)
This run confirms recent fixes on Windows: #32 (BIF crop), #33 (htj2k → ome-tiff error + --allow-nonconformant), #34 (AVIF encode), and #36 (compose BIF → cog-wsi no longer aborts on the probability associated image — this case was the deliberate expected-fail in the prior 0.26.3 run and now passes).
Performance highlights (seconds)
| Operation |
Input MB |
Sec |
Note |
| svs → dzi |
372.6 |
446.7 |
full re-tile to many files (see #25) |
| hash cog (pixel) |
630.2 |
66.9 |
decodes every htj2k L0 tile to RGB |
| cog → svs (full re-encode) |
630.2 |
53.4 |
htj2k→jpeg, whole 78k-wide pyramid |
| cog → ome-tiff jpeg (full re-encode) |
630.2 |
51.5 |
same, into OME-TIFF |
| downsample svs ×2 |
372.6 |
17.6 |
full-slide rebuild |
| hash svs (pixel) |
372.6 |
14.9 |
jpeg decode (vs 0.5s for file-hash) |
| recodec jpegxl (8k crop) |
372.6 |
9.8 |
libjxl encode |
| svs → cog-wsi (tile-copy) |
372.6 |
8.2 |
~9x svs→ome-tiff/tiff (~0.9s) for byte-identical output — see #37 |
| svs → ome-tiff / tiff (tile-copy) |
372.6 |
0.9 |
verbatim tile-copy baseline |
The timings fall into clean bands: metadata (~0.2s) < verbatim tile-copy (~1s) < rect-bounded re-encode (1.4–4.2s) < full-slide re-encode (17–53s) < full DZI re-tile (447s). AVIF encode works (transcode 4.2s), confirming #34.
Notable
Environment build: MinGW/MSYS2 on Windows 11.
Environment
A scripted sweep exercising every subcommand across the sample containers (SVS, COG-WSI, BIF) — convert across containers, crop, downsample, transcode, per-codec re-encode, composed transforms, region, extract, hash, dump-ifds, and associated-image edits. All 52 operations succeeded (0 failures).
Full per-command results (CSV) and the complete run transcript are attached:
https://gist.github.com/cornish/bb15aa58a899ef23b956ddb4fec855ca
Regression check (all green on Windows)
This run confirms recent fixes on Windows: #32 (BIF crop), #33 (htj2k → ome-tiff error +
--allow-nonconformant), #34 (AVIF encode), and #36 (compose BIF → cog-wsi no longer aborts on theprobabilityassociated image — this case was the deliberate expected-fail in the prior 0.26.3 run and now passes).Performance highlights (seconds)
The timings fall into clean bands: metadata (~0.2s) < verbatim tile-copy (~1s) < rect-bounded re-encode (1.4–4.2s) < full-slide re-encode (17–53s) < full DZI re-tile (447s). AVIF encode works (transcode 4.2s), confirming #34.
Notable
svs → cog-wsitile-copy is ~9x its tiff/ome-tiff siblings for byte-identical output — tracked in perf(investigate): svs → cog-wsi verbatim tile-copy ~10x slower than tiff/ome-tiff for identical output (IFD spooling path?) #37 (IFD spooling).doctor→Physical RAM: unknown, soft limit unlimited) — filed separately as Windows: physical-RAM detection unsupported (no ram_windows.go) — --max-memory 75% default silently inoperative #39.Environment build: MinGW/MSYS2 on Windows 11.