nanodet-plus-m-1.5x-416

Use Case and High-Level Description

The nanodet-plus-m-1.5x-416 model is one from NanoDet models family, which is a FCOS-style one-stage anchor-free object detection model which using Generalized Focal Loss as classification and regression loss. A novel label assignment strategy with a simple assign guidance module (AGM) and a dynamic soft label assigner (DSLA) is used in NanoDet-Plus to solve the optimal label assignment problem in lightweight model training. Also a light feature pyramid called Ghost-PAN is introduced in Plus models to enhance multi-layer feature fusion. The model is a super fast and high accuracy lightweight model with ShuffleNetV2 1.5x backbone. This model was pre-trained on Common Objects in Context (COCO) dataset.

More details provided in the repository.

Specification

Metric

Value

Type

Object detection

GFLOPs

3.0147

MParams

2.4614

Source framework

PyTorch*

Accuracy

Accuracy metrics obtained on Common Objects in Context (COCO) validation dataset for converted model. Label map with 80 public available object categories are used.

Metric

Value

coco_orig_precision

33.77%

coco_precision

34.53%

Input

Original model

Image, name - data, shape - 1, 3, 416, 416, format B, C, H, W, where:

  • B - batch size

  • C - number of channels

  • H - image height

  • W - image width

Expected color order is BGR.

Mean values - [103.53, 116.28, 123.675]. Scale values - [57.375, 57.12, 58.395].

Converted model

Image, name - data, shape - 1, 3, 416, 416, format B, C, H, W, where:

  • B - batch size

  • C - number of channels

  • H - image height

  • W - image width

Expected color order is BGR.

Output

Original model

The array of detection summary info, name - output, shape - 1, 3598, 112, format is B, N, 112, where:

  • B - batch size

  • N - number of detection boxes

Detection box has the following format:

  • 80 probability distribution over the classes in logits format for 80 public available Common Objects in Context (COCO) object classes, listed in file <omz_dir>/data/dataset_classes/coco_80cl.txt.

  • 8 * 4 raw coordinates in format A * 4, where A - max value of integral set.

Converted model

The array of detection summary info, name - output, shape - 1, 3598, 112, format is B, N, 112, where:

  • B - batch size

  • N - number of detection boxes

Detection box has the following format:

  • 80 probability distribution over the classes in logits format for 80 public available Common Objects in Context (COCO) object classes, listed in file <omz_dir>/data/dataset_classes/coco_80cl.txt.

  • 8 * 4 raw coordinates in format A * 4, where A - max value of integral set.

Download a Model and Convert it into OpenVINO™ IR Format

You can download models and if necessary convert them into OpenVINO™ IR format using the Model Downloader and other automation tools as shown in the examples below.

An example of using the Model Downloader:

omz_downloader --name <model_name>

An example of using the Model Converter:

omz_converter --name <model_name>

Demo usage

The model can be used in the following demos provided by the Open Model Zoo to show its capabilities: