pspnet-pytorch¶

Use Case and High-Level Description¶

pspnet-pytorch is a semantic segmentation model, pre-trained on Pascal VOC dataset for 21 object classes, listed in <omz_dir>/data/dataset_classes/voc_20cl_bkgr.txt file. The model was built on ResNetV1-50 backbone and PSP segmentation head. This model is used for pixel-level prediction tasks. For details see repository, paper.

Specification¶

Metric

Value

Type

Semantic segmentation

GFlops

357.1719

MParams

46.5827

Source framework

PyTorch*

Accuracy¶

Metric

Value

mean_iou

70.1%

Accuracy metrics were obtained with fixed input resolution 512x512.

Input¶

Original model¶

Image, name: input.1, shape: 1, 3, 512, 512, format: B, C, H, W, where:

• B - batch size

• C - number of channels

• H - image height

• W - image width

Expected color order: RGB. Mean values: [123.675, 116.28, 103.53], scale values: [58.395, 57.12, 57.375]

Converted Model¶

Image, name: input.1, shape: 1, 3, 512, 512, 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¶

Integer values in a range [0, 20], which represent an index of a predicted class for each image pixel. Name: segmentation_map, shape: 1, 1, 512, 512 in B, 1, H, W format, where:

• B - batch size

• H - image height

• W - image width

Converted Model¶

Integer values in a range [0, 20], which represent an index of a predicted class for each image pixel. Name: segmentation_map, shape: 1, 1, 512, 512 in B, 1, H, W format, where:

• B - batch size

• H - image height

• W - image width

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.

omz_downloader --name <model_name>
omz_converter --name <model_name>