Models registry#

Every model and backbone buildable from a model = dict(type=...) config block.

Build any of these from a config with model = dict(type=...), using the type string in the first column. Generated from the live registry - 53 classes (54 names including aliases), grouped by role.

Backbones#

Encoder/decoder networks used as a backbone= inside task models.

Name	Summary
`LitePT`
`MinkUNet101`
`MinkUNet14`
`MinkUNet14A`
`MinkUNet14B`
`MinkUNet14C`
`MinkUNet14D`
`MinkUNet18`
`MinkUNet18A`
`MinkUNet18B`
`MinkUNet18D`
`MinkUNet34`
`MinkUNet34A`
`MinkUNet34B`
`MinkUNet34C`
`MinkUNet50`
`PT-v2m1`
`PT-v2m2`
`PT-v2m3`
`PT-v3m1`
`PT-v3m2`
`PT-v3m4` / `PT-v3m8`	Point Transformer V3 with 3D RoPE and bottleneck CPE.
`SpUNet-v1m1`
`SpUNet-v1m2`
`SpUNet-v1m3`
`SpUNetNoSkipBase`
`Volt-v1m1`	Flat Volt encoder for Sonata pretraining.
`Volt-v1m2`	Flat Volt encoder + transposed-conv decoder, returning per-input-point features.

`pimm.models.litept.litept.LitePT`
`pimm.models.sparse_unet.mink_unet.MinkUNet101`
`pimm.models.sparse_unet.mink_unet.MinkUNet14`
`pimm.models.sparse_unet.mink_unet.MinkUNet14A`
`pimm.models.sparse_unet.mink_unet.MinkUNet14B`
`pimm.models.sparse_unet.mink_unet.MinkUNet14C`
`pimm.models.sparse_unet.mink_unet.MinkUNet14D`
`pimm.models.sparse_unet.mink_unet.MinkUNet18`
`pimm.models.sparse_unet.mink_unet.MinkUNet18A`
`pimm.models.sparse_unet.mink_unet.MinkUNet18B`
`pimm.models.sparse_unet.mink_unet.MinkUNet18D`
`pimm.models.sparse_unet.mink_unet.MinkUNet34`
`pimm.models.sparse_unet.mink_unet.MinkUNet34A`
`pimm.models.sparse_unet.mink_unet.MinkUNet34B`
`pimm.models.sparse_unet.mink_unet.MinkUNet34C`
`pimm.models.sparse_unet.mink_unet.MinkUNet50`
`pimm.models.point_transformer_v2.point_transformer_v2m1_origin.PointTransformerV2`
`pimm.models.point_transformer_v2.point_transformer_v2m2_base.PointTransformerV2`
`pimm.models.point_transformer_v2.point_transformer_v2m3_pdnorm.PointTransformerV2`
`pimm.models.point_transformer_v3.point_transformer_v3m1_base.PointTransformerV3`
`pimm.models.point_transformer_v3.point_transformer_v3m2_sonata.PointTransformerV3`
`pimm.models.point_transformer_v3.point_transformer_v3m8.PointTransformerV3`	Point Transformer V3 with 3D RoPE and bottleneck CPE.
`pimm.models.sparse_unet.spconv_unet_v1m1_base.SpUNetBase`
`pimm.models.sparse_unet.spconv_unet_v1m2_bn_momentum.SpUNetBase`
`pimm.models.sparse_unet.spconv_unet_v1m3_pdnorm.SpUNetBase`
`pimm.models.sparse_unet.spconv_unet_v1m1_base.SpUNetNoSkipBase`
`pimm.models.volt.volt_v1m1.VoltBackbone`	Flat Volt encoder for Sonata pretraining.
`pimm.models.volt.volt_v1m2.VoltBackboneV2`	Flat Volt encoder + transposed-conv decoder, returning per-input-point features.

Self-supervised pretraining#

Discriminative and masked-autoencoder pretraining systems.

Name	Summary
`PoLAr-MAE`	Self-contained PoLAr-MAE for pimm’s DefaultTrainer.
`PoLArMAE-SemSeg`	PoLAr-MAE semantic segmentation for pimm’s DefaultTrainer.
`Sonata-v1m1`
`Sonata-v1m2`

`pimm.models.polarmae.polarmae.PoLArMAE`	Self-contained PoLAr-MAE for pimm's DefaultTrainer.
`pimm.models.polarmae.polarmae_semseg.PoLArMAESemSeg`	PoLAr-MAE semantic segmentation for pimm's DefaultTrainer.
`pimm.models.sonata.sonata_v1m1_base.Sonata`
`pimm.models.sonata.sonata_v1m2_uni_teacher_head.Sonata`

Segmentation & classification#

Per-point semantic segmentation and event/point classification heads.

Name	Summary
`DefaultClassifier`
`DefaultInsSegmentor`
`DefaultSegmentor`
`DefaultSegmentorV2`
`DefaultSegmentorV3`
`DINOEnhancedSegmentor`
`PointTransformer-Cls26`
`PointTransformer-Cls38`
`PointTransformer-Cls50`
`PointTransformer-PartSeg26`
`PointTransformer-PartSeg38`
`PointTransformer-PartSeg50`
`PointTransformer-Seg26`
`PointTransformer-Seg38`
`PointTransformer-Seg50`

`pimm.models.default.DefaultClassifier`
`pimm.models.default.DefaultInsSegmentor`
`pimm.models.default.DefaultSegmentor`
`pimm.models.default.DefaultSegmentorV2`
`pimm.models.default.DefaultSegmentorV3`
`pimm.models.default.DINOEnhancedSegmentor`
`pimm.models.point_transformer.point_transformer_cls.PointTransformerCls26`
`pimm.models.point_transformer.point_transformer_cls.PointTransformerCls38`
`pimm.models.point_transformer.point_transformer_cls.PointTransformerCls50`
`pimm.models.point_transformer.point_transformer_partseg.PointTransformerSeg26`
`pimm.models.point_transformer.point_transformer_partseg.PointTransformerSeg38`
`pimm.models.point_transformer.point_transformer_partseg.PointTransformerSeg50`
`pimm.models.point_transformer.point_transformer_seg.PointTransformerSeg26`
`pimm.models.point_transformer.point_transformer_seg.PointTransformerSeg38`
`pimm.models.point_transformer.point_transformer_seg.PointTransformerSeg50`

Object detection#

Instance / panoptic detectors: Mask2Former-style Panda, autoregressive Panda, and PointGroup.

Name	Summary
`detector-v3m2`	Shared-backbone Panda detector with one batched query decoder for all labels.
`detector-v4`	Unified Panda detector: arbitrary per-query heads, optional pre-attention
`detector-v5`	Backbone, shared query decoder, losses, and panoptic postprocessing.
`detector-v5m2`	Backbone, shared query decoder, losses, and panoptic postprocessing.
`PG-v1m1`

`pimm.models.panda_detector.detector_v3m2.MultiLabelDetectorV3M2`	Shared-backbone Panda detector with one batched query decoder for all labels.
`pimm.models.panda_detector.detector_v4.UnifiedDetector`	Unified Panda detector: arbitrary per-query heads, optional pre-attention point removal, configurable queries, and per-label overlap.
`pimm.models.panda_detector.detector_v5.UnifiedDetector`	Backbone, shared query decoder, losses, and panoptic postprocessing.
`pimm.models.panda_detector.detector_v5m2.UnifiedDetector`	Backbone, shared query decoder, losses, and panoptic postprocessing.
`pimm.models.point_group.point_group_v1m1_base.PointGroup`

Adapters#

Parameter-efficient adapters that wrap an existing model.

Name	Summary
`LoRAAdapter`	Freeze a built model and LoRA-tune its attention blocks.

pimm.models.lora.LoRAAdapter

Freeze a built model and LoRA-tune its attention blocks.