Coverage Reference

Model coverage

This page documents the detector, head, backbone, and neck names exposed by the current SimpleDet registries. Coverage here means registry-backed construction metadata and the validation status shown in source and CLI discovery, not benchmark parity with any external implementation.

Use this page before selecting a detector alias or deciding whether a family is runtime validated or only compatibility routed.

Status meanings

StatusMeaning
runtime_validatedThe registry entry has native construction or tensor/runtime tests for the stated SimpleDet contract.
metadata_validatedThe alias carries structured metadata for planning and discovery, but the row itself is not a full runtime parity claim.
compatibility_aliasThe alias routes through native components, but it is not yet architecture-faithful enough to list as fully supported.
unvalidatedThe name is registered for compatibility or migration only and should not be selected as fully supported.

Planned transformer variants, historical MMDetection configs, and notebook-only references are not supported model coverage until they have a registry entry plus tensor validation.

Detector coverage

Detector rows are sourced from the native detector registry and public discovery aliases. Default detector assembly usually also needs a supported backbone; TIMM-backed defaults require simpledet[timm].

Detector Family Aliases Validation status Required extra Known limitation
RetinaNetdenseretinanet, retina, RetinaNetruntime_validatedcpuValidated for the native Retina-style dense contract, not external benchmark parity.
FCOSdensefcos, FCOSruntime_validatedcpuValidated for the native anchor-free dense contract.
ATSSdenseatss, ATSSruntime_validatedcpuValidated for the native anchor-based dense contract.
FSAFdensefsaf, FSAFcompatibility_aliascpuRouted through native dense components; not listed as architecture-faithful support.
FreeAnchordensefree_anchor, FreeAnchorcompatibility_aliascpuRetina-style compatibility path only.
GFLdensegfl, GFLruntime_validatedcpuValidated for the native GFL-style dense contract.
GFocalV2densegfocalv2, GFocalV2, GFLV2compatibility_aliascpuGFLv2 naming is routed through native dense components, but detector-level parity is not claimed.
VFNetdensevfnet, VFNetcompatibility_aliascpuUses native ATSS-compatible components with varifocal head/loss support; not fully VFNet-faithful.
FOVEA / FoveaBoxdensefovea, foveabox, FOVEA, FoveaBoxcompatibility_aliascpuNative Fovea-style head is available, but detector-level architecture parity is not claimed.
PAAdensepaa, PAAcompatibility_aliascpuProbabilistic assignment naming is exposed through a native dense route only.
RepPointsdensereppoints, RepPointscompatibility_aliascpuRequires explicit point generation settings; detector path reuses native anchor-free contracts.
YOLOFdenseyolof, YOLOFcompatibility_aliascpuYOLOF-style head and DilatedEncoder neck exist; detector-level parity is not claimed.
TOODdensetood, TOODcompatibility_aliascpuTask-aligned dense naming route only.
DDODdenseddod, DDODcompatibility_aliascpuDecoupled dense naming route only.
AutoAssigndenseauto_assign, AutoAssigncompatibility_aliascpuAuto-assignment naming route only.
NAS-FCOSdensenas_fcos, NAS-FCOScompatibility_aliascpuNAS-FCOS naming route only.
CenterNetdensecenternet, CenterNetruntime_validatedcpuValidated for native heatmap, width-height, offset, and decode paths.
CornerNetdensecornernet, CornerNetruntime_validatedcpuValidated for paired corner heatmaps, embeddings, offsets, and native decode.
YOLO familydenseyolo, YOLO, yolo3, yolo4, yolo_v3, yolov3, yolov4, yolov5, yolov6, yolov7, yolov8, yolov9, yolov10compatibility_aliascpuRoutes through the YOLOX-compatible neck, head, target, and decode path; version-specific YOLO parity is not claimed.
YOLOXdenseyolox, YOLOXruntime_validatedcpuValidated with YOLOXPAFPN, objectness outputs, simOTA-style targets, and decode.
RTMDetdensertmdet, rtmdet_tiny, rtmdet_s, rtmdet_m, rtmdet_l, RTMDetruntime_validatedcpuValidated for the native RTMDet head and YOLOXPAFPN-compatible path.
SSDdensessd, ssd300, ssd512, ssdlite, SSDruntime_validatedcpuValidated for SSDNeck, native SSD head, and anchor decode.
EfficientDetdenseefficientdet, efficientdet_d0, efficientdet_d1, efficientdet_d2, EfficientDetruntime_validatedcpu; default backbone uses timmValidated for EfficientNet/BiFPN/native head assembly; install simpledet[timm] for default native module construction.
SOLOv2densesolov2, solov2_lightcompatibility_aliascpuHistorical dense route only; instance segmentation parity is not claimed.
DETRtransformerdetr, DETRruntime_validatedcpu2D query detector scaffold with native logits, boxes, and decode.
Conditional DETRtransformerconditional_detr, conditionaldetr, ConditionalDETR, Conditional DETRruntime_validatedcpu2D query detector scaffold; unsupported v2 names are rejected.
DAB-DETRtransformerdab_detr, dabdetr, DAB-DETR, DABDETRruntime_validatedcpu2D query detector scaffold; unsupported v2 names are rejected.
Deformable DETRtransformerdeformable_detr, deformabledetr, DeformableDETR, Deformable DETRruntime_validatedcpu2D query detector scaffold; Deformable DETR v2 names are planned, unsupported.
DINOtransformerdino, DINOruntime_validatedcpu2D query detector scaffold; DINOv2 is planned, unsupported.
RPNproposalrpn, RPN, RPN detector, Region Proposal Networkruntime_validatedcpuProposal-only detector path.
Fast R-CNNroifast_rcnn, fast-rcnn, Fast R-CNNruntime_validatedcpuROI bbox path without an RPN stage.
Faster R-CNNroifaster_rcnn, faster-rcnn, Faster R-CNNruntime_validatedcpuNative RPN plus ROI bbox path.
Mask R-CNNroimask_rcnn, mask-rcnn, Mask R-CNNruntime_validatedcpuNative mask head contract is validated; full dataset benchmark parity is not claimed.
Grid R-CNNroigrid_rcnn, gridrcnn, Grid R-CNNruntime_validatedcpuNative grid ROI head contract is validated.
Cascade R-CNNroicascade_rcnn, cascadercnn, Cascade R-CNNruntime_validatedcpuNative cascade bbox refinement path is validated.
Cascade Mask R-CNNroicascade_mask_rcnn, cascademaskrcnn, Cascade Mask R-CNNruntime_validatedcpuNative cascade bbox plus mask contracts are validated.
Libra R-CNNroilibra_rcnn, librarcnn, Libra R-CNNruntime_validatedcpuBalanced ROI sampling contract is represented; external parity is not claimed.
Double-Head R-CNNroidouble_head_rcnn, doubleheadrcnn, Double-Head R-CNNruntime_validatedcpuNative double-head ROI feature contract is validated.
Dynamic R-CNNroidynamic_rcnn, dynamicrcnn, Dynamic R-CNNruntime_validatedcpuNative dynamic ROI feature path is validated.
Sparse R-CNNroisparse_rcnn, sparsercnn, Sparse R-CNNruntime_validatedcpuValidated with learned proposal boxes/features and sparse ROI refinement.

Head coverage

Head rows are registry entries. Rows marked unvalidated are compatibility names and must not be treated as fully supported heads.

HeadKindAliasesValidation statusRequired extraKnown limitation
RetinaHeaddenseretina, retina_headruntime_validatedcpuWraps the native Retina-style dense output contract.
RetinaNetHeadhead-unvalidated-Compatibility name only; use RetinaHead for validated support.
FCOSHeaddensefcos, fcos_headruntime_validatedcpuNative anchor-free dense contract.
FCOSV2Headhead-unvalidated-Registered naming variant only.
FCOSHeadV2head-unvalidated-Registered naming variant only.
FSAFHeaddensefsaf, fsaf_headruntime_validatedcpuFSAF-style head contract, separate from detector-level architecture parity.
FreeAnchorRetinaHeaddensefree_anchor, free_anchor_head, free_anchor_retina_headruntime_validatedcpuRetina-style dense head variant.
GFLHeaddensegfl, gfl_head, gfocal, gfocal_headruntime_validatedcpuUses the native ATSS-style dense output contract.
GFLV2HeaddenseGFLV2, gflv2_head, gfocalv2, gfocalv2_headruntime_validatedcpuHead contract is validated; detector-level GFocalV2 remains a compatibility alias.
ATSSHeaddenseatss, atss_headruntime_validatedcpuNative anchor-based dense contract.
ATSSV2Headhead-unvalidated-Registered naming variant only.
RPNHeaddenserpn, rpn_headruntime_validatedcpuClass-agnostic proposal head.
VFNetHeaddenseVFNet, vfnet_headruntime_validatedcpuHead contract is validated; detector-level VFNet remains a compatibility alias.
RepPointsHeaddenseRepPoints, reppoints_headruntime_validatedcpuRequires point stride metadata at construction.
FoveaHeaddenseFOVEA, fovea_headruntime_validatedcpuHead contract is validated; detector-level FoveaBox remains a compatibility alias.
PAAHeaddensePAA, paa_headruntime_validatedcpuUses the native ATSS-style dense output contract.
TOODHeaddenseTOOD, tood_headruntime_validatedcpuUses the native FCOS-style dense output contract.
DDODHeaddenseDDOD, ddod_headruntime_validatedcpuUses the native ATSS-style dense output contract.
AutoAssignHeaddenseAutoAssign, auto_assign_headruntime_validatedcpuUses the native FCOS-style dense output contract.
NASFCOSHeaddenseNAS-FCOS, nas_fcos_headruntime_validatedcpuUses the native FCOS-style dense output contract.
YOLOFHeaddenseYOLOF, yolof_headruntime_validatedcpuHead contract is validated; detector-level YOLOF remains a compatibility alias.
YOLOXHeaddenseyolox, yolox_headruntime_validatedcpuIncludes explicit objectness, class, and bbox branches.
YOLOHeadhead-unvalidated-Generic YOLO name only; use validated family-specific rows.
RTMDetHeaddensertmdet, rtmdet_headruntime_validatedcpuNative RTMDet class and bbox branches.
SSDHeaddensessd, ssd_head, ssd300, ssd512, ssdliteruntime_validatedcpuAnchor-based SSD dense contract.
EfficientDetHeaddenseefficientdet, efficientdet_headruntime_validatedcpuAnchor-based EfficientDet head contract.
CenterNetHeaddenseCenterNet, centernet_headruntime_validatedcpuKeypoint heatmap head with dense compatibility outputs.
CornerNetHeaddenseCornerNet, cornernet_headruntime_validatedcpuPaired-corner heatmap head with embeddings and offsets.
SOLOV2Headhead-unvalidated-Registered name only; no supported instance segmentation claim.
Shared2FCBBoxHeadroishared_2fc_bbox_head, shared2fcruntime_validatedcpuCommon two-FC ROI bbox head.
ConvFCBBoxHeadroiconvfc_bbox_head, convfcruntime_validatedcpuConfigurable conv plus FC ROI bbox head.
DoubleConvFCBBoxHeadroidouble_convfc_bbox_headruntime_validatedcpuDouble-head ROI bbox contract.
DynamicBBoxHeadroidynamic_bbox_head, dynamic_bboxruntime_validatedcpuDynamic proposal feature mixing contract.
CascadeBBoxHeadroicascade_bbox_head, cascade_bboxruntime_validatedcpuCascade stage refinement contract.
SABLHeadroisabl_head, sabl_bbox_head, side_aware_bbox_headruntime_validatedcpuSide-aware boundary localization ROI head.
SparseRoIHeadroisparse_roi_head, sparse_bbox_head, sparse_rcnn_roi_headruntime_validatedcpuSparse R-CNN style learned proposal feature head.
FCNMaskHeadroifcn_mask_head, fcn_mask, mask_headruntime_validatedcpuFCN-style ROI mask head.
CascadeMaskHeadroicascade_mask_head, cascade_maskruntime_validatedcpuCascade mask head contract.
GridHeadroigrid_head, grid_roi_headruntime_validatedcpuGrid R-CNN ROI grid contract.
DETRHeadtransformerDETR, detr_headruntime_validatedcpuQuery class and box predictions.
ConditionalDETRHeadtransformerConditionalDETR, conditional_detr_headruntime_validatedcpuConditional query content variant.
DABDETRHeadtransformerDAB-DETR, dab_detr_headruntime_validatedcpuDynamic anchor box query variant.
DeformableDETRHeadtransformerDeformableDETR, deformable_detr_headruntime_validatedcpuMulti-scale transformer head with level embeddings.
DINOHeadtransformerDINO, dino_headruntime_validatedcpuAnchor and denoising query embeddings.

Backbone coverage

Native backbone aliases are planning metadata for TIMM-backed feature extraction. Install simpledet[cpu,timm] for runtime construction.

BackboneFamilyAliasesValidation statusRequired extraKnown limitation
timmbackboneTimmEncoder, timm:<model>runtime_validatedcpu, timmRequires TIMM feature metadata at runtime.
resnet18ResNetResNet-18, r18metadata_validatedcpu, timmAlias carries four-stage channel metadata.
resnet50ResNetResNet, ResNet-50, r50metadata_validatedcpu, timmAlias carries four-stage channel metadata.
resnet101ResNetResNet-101, r101metadata_validatedcpu, timmAlias carries four-stage channel metadata.
resnext50_32x4dResNeXtResNeXt, ResNeXt-50metadata_validatedcpu, timmAlias carries four-stage channel metadata.
res2net50_26w_4sRes2NetRes2Net, Res2Net-50metadata_validatedcpu, timmAlias carries four-stage channel metadata.
hrnet_w18HRNetHRNet, HRNet-W18, hrnet18metadata_validatedcpu, timmAlias carries multi-resolution stage metadata.
cspdarknet53CSPDarkNetCSPDarkNet, CSPDarkNet-53, darknet53metadata_validatedcpu, timmAlias carries four-stage channel metadata.
cspnext_tinyCSPNeXtCSPNeXt, CSPNeXt-Tinymetadata_validatedcpu, timmAlias carries four-stage channel metadata.
mobilenetv2_100MobileNetV2MobileNetV2metadata_validatedcpu, timmAlias carries four-stage channel metadata.
mobilenetv3_large_100MobileNetV3MobileNetV3, MobileNetV3-Largemetadata_validatedcpu, timmAlias carries four-stage channel metadata.
efficientnet_b0EfficientNetEfficientNet, EfficientNet-B0metadata_validatedcpu, timmDefault EfficientDet backbone alias.
convnext_tinyConvNeXtConvNeXt, ConvNeXt-Tinyruntime_validatedcpuNative ConvNeXt-Tiny backbone built from reusable CNN blocks.
swin_tiny_patch4_window7_224Swin TransformerSwin, Swin Transformer, Swin-T, swin_tinymetadata_validatedcpu, timmAlias carries four-stage channel metadata.
vit_base_patch16_224Vision TransformerViT, Vision Transformer, ViT-B/16, vit_basemetadata_validatedcpu, timmViT stages share the configured channel width in current metadata.

Neck coverage

NeckKindAliasesValidation statusRequired extraKnown limitation
FPNneckfpnruntime_validatedcpuRequires feature sequence channels to match the compiled plan.
ChannelMapperneckchannel_mapperruntime_validatedcpuProjection-only neck for multiscale features.
PANneck-runtime_validatedcpuPath aggregation FPN variant.
PANETneckpanet_neckruntime_validatedcpuPANet naming variant.
BiFPNneckbi_fpnruntime_validatedcpuBidirectional feature pyramid variant.
BiFPNV2neckbi_fpn_v2runtime_validatedcpuBiFPNv2 naming variant.
PAFPNneckpa_fpnruntime_validatedcpuPath aggregation FPN naming used by dense detectors.
NASFPNnecknas_fpnruntime_validatedcpuRepeated adjacent-level fusion; not a NAS search implementation.
FPNLiteneck-runtime_validatedcpuLightweight FPN naming variant.
FPNLiteNeckneck-runtime_validatedcpuCamel-case alias for lightweight FPN callers.
DilatedEncoderneckdilated_encoderruntime_validatedcpuYOLOF-style single-feature-level contract.
HRFPNneckhr_fpnruntime_validatedcpuHRNet-style multi-resolution feature fusion.
SSDNeckneckssd_neckruntime_validatedcpuSSD projection and extra-level neck.
YOLOXPAFPNneckyolox_pafpnruntime_validatedcpuYOLOX path aggregation feature pyramid contract.

Not supported yet

  • DETR3D, DETRv2, Deformable DETR v2, Conditional DETR v2, DAB-DETR v2, and DINOv2 are planned, unsupported transformer variants. The builders reject them instead of silently routing them to another detector.
  • Legacy MMDetection config files under MyConfigs/, old notebooks, and historical study scripts are config-only references. They are not SimpleDet support evidence.
  • Rows marked compatibility_alias or unvalidated must not be presented as fully supported architecture-faithful implementations.