mask_rcnn_inception_resnet_v2_atrous_coco

Use Case and High-Level Description

Mask R-CNN Inception ResNet V2 Atrous is trained on Common Objects in Context (COCO) dataset and used for object instance segmentation. For details, see a paper.

Specification

Metric

Value

Type

Instance segmentation

GFlops

675.314

MParams

92.368

Source framework

TensorFlow*

Accuracy

Metric

Value

coco_orig_precision

39.86%

coco_orig_segm_precision

35.36%

Input

Original Model

Image, name: image_tensor, shape: 1, 800, 1365, 3, format: B, H, W, C, where:

  • B - batch size

  • H - image height

  • W - image width

  • C - number of channels

Expected color order: RGB.

Converted Model

  1. Image, name: image_tensor, shape: 1, 800, 1365, 3, format: B, H, W, C, where:

    • B - batch size

    • H - image height

    • W - image width

    • C - number of channels

    Expected color order: BGR.

  2. Information of input image size, name: image_info, shape: 1, 3, format: B, C, where:

    • B - batch size

    • C - vector of 3 values in format H, W, S, where H is an image height, W is an image width, S is an image scale factor (usually 1)

Output

Original Model

  1. Classifier, name: detection_classes. Contains predicted bounding boxes classes in a range [1, 91]. The model was trained on Common Objects in Context (COCO) dataset version with 90 categories of objects, 0 class is for background.

  2. Probability, name: detection_scores. Contains probability of detected bounding boxes.

  3. Detection box, name: detection_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 rescaled to input image size.

  4. Detections number, name: num_detections. Contains the number of predicted detection boxes.

  5. Segmentation mask, name: detection_masks. Contains segmentation heatmaps of detected objects for all classes for every output bounding box.

Converted Model

  1. The array of summary detection information, name: reshape_do_2d, shape: 100, 7 in the format N, 7, where N is the number of detected bounding boxes. For each detection, the description has the format: [image_id, label, conf, x_min, y_min, x_max, y_max], where:

    • image_id - ID of the image in the batch

    • label - predicted class ID

    • conf - confidence for the predicted class

    • (x_min, y_min) - coordinates of the top left bounding box corner (coordinates stored in normalized format, in range [0, 1])

    • (x_max, y_max) - coordinates of the bottom right bounding box corner (coordinates stored in normalized format, in range [0, 1])

  2. Segmentation heatmaps for all classes for every output bounding box, name: masks, shape: 100, 90, 33, 33 in the format N, 90, 33, 33, where N is the number of detected masks, 90 is the number of classes (the background class excluded).

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: