After a successful launch a couple of weeks ago, we're back with our second release! With a few exceptions this release focuses on internal refactors & improving our processes. However, we think you'll enjoy these goodies that made it in nonetheless!

https://user-images.githubusercontent.com/2910679/222510504-23871b8c-0bef-49c2-bbd2-37baab4247e8.mp4

You can now generate point clouds directly from depth textures and choose a wide variety of color maps. Check out this video on how to use it. This is a lot faster and more convenient than doing so manually in your own code Some caveats: Picking is not yet working and visible history may behave differently (related to #723)

Other highlights:

• Viewer
  • Improved formatting of date-times in plots #1356
  • Labels for 3D objects have now a color can now be selected & hovered #1438
  • Scale factor is saved across sessions and more persistent between screens #1448
  • Showing tensors in the viewer is now faster
• SDK
  • Python packages now work with Ubuntu-20.04 #1334
  • u8 segmentation stay u8 now (they converted to u16 before) #1376
  • 2D Line strips can now be logged directly #1430
  • Add a strict mode to the Python SDK where misuses of the API result in exceptions being raised.#1477
  • Fix disabling Python API through init not working #1517
• General
  • We build now with fewer build dependencies (there is however still more work to do!).
    Notably, we previously used a version of the time crate which had a security issue (CVE-2020-26235), thanks @mpizenberg for helping out!
  • Print more information & troubleshooting info on crash

Meanwhile, we did a bunch of improvements to our manual. If you had trouble running Rerun so far, check our updated troubleshooting page (and as always, please open an issue if something doesn't work).

⚠️ BREAKING: old .rrd files no longer load ⚠️

In Detail

New Features

• Generate point clouds directly from depth textures
  • re_renderer: implement depth cloud renderer #1415
  • Integrate depth clouds into Rerun #1421
  • CPU & GPU color maps #1484
  • Integrate GPU color maps into depth clouds #1486
• Python SDK: Add strict mode #1477
• OS independent Zoom factor & serialization thereof #1448
• Labels for 3D objects have now a color can now be selected & hovered #1438
• Add 2d support for linestrips #1430
• Add signal handler on *nix with troubleshooting and stacktrace #1340
  • Point users to our troubleshooting page on panic #1338

Performance

• Speed up conversions for color arrays in Python #1454
• Speed up fixed-sized array iteration #1050
• Speed up tensor handling by padding data through more directly
  • Direct conversion to dynamic image from Tensors #1455
  • Convert view_tensor to use the new native Tensors #1439
• Add option to show performance metrics in the UI in release builds too #1444
• Faster stable diffusion sample #1364
• SDK: stream to disk with save feature #1405
• re_renderer has now a direct CPU->GPU copy mechanism
  • CpuWriteGpuReadBelt for fast frame by frame memory transfers #1382
  • Uniform buffer utility using CpuWriteGpuReadBelt #1400
  • Use CpuWriteGpuReadBelt for mesh data gpu upload #1416

Small improvements & Bugfixes

• UI
  • Add scroll-bars the "Add/Remove entities" window #1445
  • Unify the time formatting between the time panel and the plot #1369
  • Timeline
    • Fix precision issue when zooming in on the timeline #1370
    • Improve the gap-detector #1363
  • Better time axis on plot view #1356
  • Prevent wrap on 'Streams' text #1308
  • Update to eframe 0.21.3 with fix for web text input #1311
• re_renderer
  • Fix crash due to always expecting Rgba8Unorm backbuffer on Web & Bgra8Unorm on native #1413
  • Allow controlling the graphics backend & power preference through standard wgpu env vars #1332
• Heuristic for camera frustum length is now based on scene size #1433
• Fix python type signature for tensor names #1443
• Don't convert u8 segmentation images to u16 #1376
• Docs (excluding the manual)
  • Improve the docs of connect and serve #1450
  • Update log_mesh and log_meshes docs. #1286
  • Add guidelines for adding dependencies in a PR #1431
  • Add a few more sections to CODE_STYLE.md #1365
  • Fixup for some doc links #1314
  • Document undocumented environment variables on help page. #1335
  • Link to SDK operating modes doc in both SDK #1330
• More information in --version #1388
• Remove already broken show method from Python SDK #1429
• Analytics
  • Send analytics events with callstacks on panics and signals #1409
  • Put all analytics to one bucket #1390
  • add event for when we serve the web-viewer .wasm #1379
  • register SDK language and data source #1371
  • Refactor analytics #1368
• Versioned log streams streams #1420
• Fix path issues when running debug viewer within workspace #1341
• Detailed errors for re_renderer include_file! #1339
• Limit logging in web-viewer to warn in order to workaround a crash issue (and reduce log spam) 1514
• Fix disabling API through init not working #1517

CI, Testing & Build improvements

• Reduce build dependencies
  • Get rid of time 0.1.* dependency #1408
  • Remove unnecessary ordered-float #1461
  • Remove extraneous image features and dependencies #1425
  • Replace reqwest with ureq #1407
  • Remove derive_more dependency #1406
• Use different artifact names for wasm/js in debug builds #1428
• Separate mac wheels & trigger wheel build from ui #1499
• Add spell checking to CI #1492
• Repo size
  • Always create new orphaned branch for gh-pages #1490
  • GitHub Action to prevent large files #1478
• Python
  • Remove the python job path filters #1452
  • Use ruff for our python lints #1378
  • Use python3 in the jobs that weren't tested in PR #1348
• Testing
  • Add a test of memory use when logging a lot of big images #1372
• Switch ci_docker to a container based on ubuntu 20.04 #1334
• Release handling
  • Switch release action to ncipollo #1489
  • Fix our continuous pre-releases #1458
  • Delete the prerelease before creating the new one #1485
  • Set prerelease to true even for version-tagged CI job #1504
  • Let the release job take care of creating the tag #1501
  • Use cargo update -w instead of cargo check when prepping prerelease #1500
  • Use prerelease tag instead of latest and update pointer on prerelease #1481
  • Include date in pre-release version #1472
  • Switch pre-release action to ncipollo/release-action #1466
• Disallow some methods and types via Clippy#1411

Other not user facing refactors

• Fix: don't create a dummy LogDb when opening the Rerun Menu #1440
• re_renderer
  • Draw Phases in preparation of executing Renderer several times on different targets #1419
    • Fix mesh creation failing to copy index data. #1473
    • do not silently drop draw phases #1471
  • Simplify bind group allocation call by passing pool collection object. #1459
  • Interior mutable buffer/texture/bindgroup pools #1374
  • Rename all instances of frame_maintenance to begin_frame #1360
  • Texture & buffer call now wgpu's destroy on removal from pool #1359
  • Arrow buffers as (optional) first-class citizen #1482
  • Log static re_renderer resource generation #1464
• Internal log_text_entry_internal to break circular deps #1488
• Delete ClassicTensor and cleanup #1456
• Fix re_renderer file watcher watching the same file several times #1463
• Analytics
  • More ergonomic API #1410
  • Streamlining host vs. recorder python/rust versions #1380
  • Fix workspace detection #1437
• Introduce DeserializableComponent trait and high-level query_latest #1417

Full Changelog