NanoDet
Category: model.detection.2d
Version: 1.0.0
UUID: 70be5afb982ce326322353d3f19a73683861c06c8db1a5aec88a86b9bded8ee3
Execution context: leip_af
Choice Priority: 0.3
Subcomponent Parameters
Value Parameters
| Name |
Synonyms |
Type |
Values |
| Input Height |
model.input.height |
scalar |
int, min: 1 |
| Input Width |
model.input.width |
scalar |
int, min: 1 |
| Model Architecture |
model.variation,
model.backbone,
model.architecture |
choice |
10 choices |
| IoU Threshold for NMS |
post_processor.iou_threshold |
scalar |
float, min: 0.0, max: 1.0 |
| Loss weight of GIoULoss |
model.loss.giou_loss.weight |
scalar |
float |
| Loss weight of DistributionFocalLoss |
model.loss.distribution_focal_loss.weight |
scalar |
float |
| The beta parameter for calculating the modulating factor in QualityFocalLoss |
model.loss.quality_focal_loss.beta |
scalar |
float |
| Loss weight of QualityFocalLoss |
model.loss.quality_focal_loss.weight |
scalar |
float |
| Max number of detections per sample |
post_processor.max_detections |
scalar |
int, min: 0 |
| Use Pretrained Weights |
model.use_pretrained |
scalar |
bool |
| Use Pretrained Backbone |
model.backbone.use_pretrained |
scalar |
bool |
| Pretrained Variant |
model.weights.pretrained_variant |
choice |
320
416
512 |
| Prediction Confidence Threshold |
post_processor.confidence,
post_processor.confidence_threshold |
scalar |
float, min: 0.0, max: 1.0, step: 0.05 |
| Include decoding in Graph |
model.include_decoding,
export.include_decoding |
scalar |
bool |
Constraints
- NanoDet input sizes have to be multiples of 32 (
model.input.width%32==0 and model.input.height%32==0)
This component fits into
Model Architecture
nanodet-m (NanoDet-m)
nanodet-plus-m (NanoDet-Plus-m)
nanodet-plus-m-1.5x (NanoDet-Plus-m-1.5x)
nanodet-m-1.5x (NanoDet-m-1.5x)
nanodet-t (NanoDet-t)
nanodet-g (NanoDet-g)
nanodet-efficient-lite0 (NanoDet-Efficient-Lite0)
nanodet-efficient-lite1 (NanoDet-Efficient-Lite1)
nanodet-efficient-lite2 (NanoDet-Efficient-Lite2)
nanodet-repvgg (NanoDet-RepVGG)