Repository navigation
Releases
Releases · arthurdjn/pytorch-pointcloud
v0.0.22
What's Changed
- feat: add group-free 3d detector by @arthurdjn in #481
- fix: check registered weight files through the hub api instead of download requests by @arthurdjn in #497
- fix: correct pooling, serialization and neighbor ops on edge inputs by @arthurdjn in #499
- fix: harden dataset caching, worker seeds and registry reads by @arthurdjn in #500
- fix: correct transforms on integer, swapped and seeded inputs by @arthurdjn in #501
- fix: correct detection losses and box ops by @arthurdjn in #502
- fix: follow the nuScenes evaluation protocol by @arthurdjn in #503
- fix: align pointnext heads and configs, point mamba dropout by @arthurdjn in #504
- fix: align registered point model inputs with the weights by @arthurdjn in #505
- fix: pick best checkpoints on validation and repair example scripts by @arthurdjn in #506
- chore: register the metrics measured on the benchmark rerun by @arthurdjn in #507
- fix: wait for the hub api quota in the hub sync test by @arthurdjn in #509
- chore: bump version 0.0.21 -> 0.0.22 by @arthurdjn in #510
Full Changelog: v0.0.21...v0.0.22
v0.0.21
What's Changed
- ci: keep the runs of every commit on main by @arthurdjn in #495
- feat: add minkunet models by @arthurdjn in #474
- feat: add cylinder3d models by @arthurdjn in #478
- feat: add fcaf3d and tr3d detectors by @arthurdjn in #479
- feat: add voxel r-cnn detector by @arthurdjn in #482
- feat: add pct and curvenet models by @arthurdjn in #483
- chore: bump version 0.0.20 -> 0.0.21 by @arthurdjn in #496
Full Changelog: v0.0.20...v0.0.21
v0.0.20
What's Changed
- feat: add pointnet pretrained models by @arthurdjn in #469
- feat: add litept models by @arthurdjn in #476
- feat: add volt models by @arthurdjn in #477
- feat: add recon, act, point2vec and pcp-mae models by @arthurdjn in #484
- docs: changelog convention and weight migration steps in the contributing guide by @arthurdjn in #420
- fix: wrong results of layers and ops on unsorted batches and edge cases by @arthurdjn in #488
- fix: degenerate-box IoU and detection head edge cases by @arthurdjn in #490
- fix: metric and inferer edge cases by @arthurdjn in #491
- fix: dtype and edge-case bugs in transforms and datasets by @arthurdjn in #492
- fix: head, dropout and transform bugs in registered models by @arthurdjn in #489
- docs: changelog note for the point-m2ae pretraining mask change by @arthurdjn in #493
- chore: bump version 0.0.20 by @arthurdjn in #494
Full Changelog: v0.0.19...v0.0.20
v0.0.19
What's Changed
- chore: remove deprecated model names by @arthurdjn in #458
- feat: add nuscnes lidarseg and ptv3 model by @arthurdjn in #452
- refactor: add functional voxelize and voxel grid origin by @arthurdjn in #460
- feat: add myria3d model and associated benchmark by @arthurdjn in #453
- feat: add mmdet3d centerpoint models and layers by @arthurdjn in #459
- feat: version every processed dataset cache by @arthurdjn in #461
- feat: add mmdet3d part-a2 and pv-rcnn models by @arthurdjn in #465
- refactor: expose the center losses on plain tensors by @arthurdjn in #467
- test: check hub cards, configs and pins against the registry by @arthurdjn in #468
- feat: add ptv3 gridnet-hd checkpoint and dataset by @arthurdjn in #464
- feat: add transfusion-l detector by @arthurdjn in #480
- feat: add mmdet3d, openpcdet and 3detr detector checkpoints with second-iou and pointpillars-fpn by @arthurdjn in #472
- feat: add pointgpt part segmentation models by @arthurdjn in #470
- feat: add pvcnn and pointnet++ shapenetpart models by @arthurdjn in #471
- chore: bump version 0.0.18 -> 0.0.19 by @arthurdjn in #487
Full Changelog: v0.0.18...v0.0.19
v0.0.18
What's Changed
- fix: ocnn octree cache by @arthurdjn in #454
- refactor: deprecated model decorator by @arthurdjn in #455
- fix: open3d evaluation protocol restoration by @arthurdjn in #456
- chore: bump version 0.0.17 -> 0.0.18 by @arthurdjn in #457
Full Changelog: v0.0.17...v0.0.18
v0.0.17
What's Changed
- test: re-upload non-regression tests for models that changed output shape by @arthurdjn in #444
- fix: batch norm eps for ptv3 pretrained weights by @arthurdjn in #445
- feat: open3d model zoo by @arthurdjn in #431
- feat: add more size variants to foundation models with filter fn by @arthurdjn in #446
- feat: add ptv3 trained on dales dataset by @arthurdjn in #448
- feat: pvcnn and pvcnn2 mit-han-lab pretrained models by @arthurdjn in #449
- chore: bump version 0.0.16 -> 0.0.17 by @arthurdjn in #450
- fix: pvcnn revision weights by @arthurdjn in #451
Full Changelog: v0.0.16...v0.0.17
v0.0.16
What's Changed
- docs: add scores chart and improve table readability by @arthurdjn in #441
- chore: bump version 0.0.15 -> 0.0.16 by @arthurdjn in #442
Full Changelog: v0.0.15...v0.0.16
v0.0.15
What's Changed
- fix: torch emoji render by @arthurdjn in #433
- fix: syntax warning on first import under python 3.12 by @arthurdjn in #435
- feat: read and write ply files with numpy, drop plyfile by @arthurdjn in #437
- docs: fix stale notice paths, the num_classes faq entry and the model zoo notebooks by @arthurdjn in #438
- chore: bump version 0.0.14 -> 0.0.15 by @arthurdjn in #440
Full Changelog: v0.0.14...v0.0.15
v0.0.14
What's Changed
- fix: dgcnn lines of benchmark.sh and a parse test for every line by @arthurdjn in #427
- fix: consistent layer factories, box columns and param groups by @arthurdjn in #426
- fix: actionable missing-dependency errors from create_model by @arthurdjn in #422
- chore: drop TORCH_POINTCLOUD_MODELS_DIR, weights live under the cache by @arthurdjn in #430
- fix: rebuild the head after loading pretrained weights when num_classes is overridden by @arthurdjn in #425
- feat: warn on non-commercial pretrained weights by @arthurdjn in #423
- fix: runnable docstring examples and a test over every fence by @arthurdjn in #424
- feat: explicit public exports and an ops namespace by @arthurdjn in #421
- docs: data format, faq and about pages and project nav by @arthurdjn in #428
- chore: bump version 0.0.13 -> 0.0.14 by @arthurdjn in #432
Full Changelog: v0.0.13...v0.0.14
v0.0.13
What's Changed
- refactor: pretrain model outputs by @arthurdjn in #392
- feat: improved transforms and utils repeat dataset by @arthurdjn in #393
- feat: optim package with reusable tools by @arthurdjn in #394
- feat: model_info registry accessor by @arthurdjn in #396
- feat: detection head initialization and velocity targets by @arthurdjn in #397
- feat: box transforms for the detection recipes by @arthurdjn in #398
- refactor: one example directory per model by @arthurdjn in #399
- feat: classification and part segmentation recipes by @arthurdjn in #400
- feat: semantic segmentation recipes by @arthurdjn in #405
- feat: finetuning recipes of the masked point modeling models by @arthurdjn in #406
- feat: detection recipes by @arthurdjn in #407
- docs: simplify banner as a single PNG by @arthurdjn in #409
- fix: models references and better tests error message by @arthurdjn in #410
- docs: light theme banner by @arthurdjn in #411
- docs: update README with better spacing by @arthurdjn in #412
- docs: better grids banner for homepage with sidenav by @arthurdjn in #413
- docs: transparent banner grid by @arthurdjn in #414
- docs: banner border color and spacing by @arthurdjn in #415
- docs: fix missing figure by @arthurdjn in #416
- chore: format constants by @arthurdjn in #417
- format: better spacing in readme between sections by @arthurdjn in #418
- chore: bump version 0.0.12 -> 0.0.13 by @arthurdjn in #419
Full Changelog: v0.0.12...v0.0.13