yolact-resnet50-fpn-pytorch

Use Case and High-Level Description

YOLACT ResNet 50 is a simple, fully convolutional model for real-time instance segmentation described in “YOLACT: Real-time Instance Segmentation” paper. Model pre-trained in Pytorch* on Common Objects in Context (COCO) dataset. For details, see the repository.

Specification

Metric

Value

Type

Instance segmentation

GFlops

118.575

MParams

36.829

Source framework

PyTorch*

Accuracy

Metric

Value

AP@masks

28.00%

AP@boxes

30.69%

Input

Original Model

Image, name: input.1, shape: 1, 3, 550, 550, format: B, C, H, W, where:

  • B - batch size

  • H - image height

  • W - image width

  • C - number of channels

Expected color order: RGB. Mean values - [123.675, 116.78, 103.94], scale values - [58.395, 57.12, 57.375].

Converted Model

Image, name: input.1, shape: 1, 3, 550, 550, format: B, C, H, W, where:

  • B - batch size

  • C - number of channels

  • H - image height

  • W - image width

Expected color order: BGR.

Output

Original Model

  1. Detection scores, name: conf. Contains score distribution over all classes in the [0,1] range . The model was trained on Common Objects in Context (COCO) dataset version with 80 categories of objects, 0 class is for background. Output shape is 1, 19248, 81 in B, N, C format, where:

    • B - batch size,

    • N - number of detected boxes,

    • C - number of classes.

  2. Detection boxes, name: boxes. Contains detection boxes coordinates in a format [y_min, x_min, y_max, x_max], where (x_min, y_min) are coordinates of the top left corner, (x_max, y_max) are coordinates of the right bottom corner. Coordinates are normalized in [0, 1] range. Output shape is 1, 19248, 4 in B, N, 4 format, where:

    • B - batch size,

    • N - number of detected boxes.

  3. Masks features prototypes, name: proto. Contains the features projection for instance mask decoding. Output shape is 1, 138, 138, 32 in B, H, W, C, where:

    • B - batch size,

    • H - mask height,

    • W - mask width,

    • C - channels.

  4. Raw instance masks, name: mask. Contains segmentation heatmaps of detected objects for all classes for every output bounding box. Output shape is B, N, C format, where:

    • B - batch size,

    • N - number of detected boxes,

    • C - channels.

Final mask prediction can be obtained by matrix multiplication of proto and transposed mask outputs.

Converted Model

Converted model outputs are the same as in the original model.

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: