This is a vehicle reidentification model for a general scenario. It uses a whole car body image as an input and outputs an embedding vector to match a pair of images by the cosine distance. The model is based on the OmniScaleNet backbone developed for fast inference. A single reidentification head from the 1/16 scale feature map outputs an embedding vector of 512 floats.
Metric | Value |
---|---|
VeRi-776* rank-1 | 96.31 % |
VeRi-776* mAP | 85.15 % |
Camera location | All traffic cameras |
Support of occluded vehicles | YES |
Occlusion coverage | <50% |
GFlops | 2.643 |
MParams | 2.183 |
Source framework | PyTorch* |
The cumulative matching curve (CMC) at rank-1 is accuracy denoting the possibility to locate at least one true positive in the top-1 rank. Mean Average Precision (mAP) is the mean across Average Precision (AP) of all queries. AP is defined as the area under the precision and recall curve.
One image of the shape [1x3x208x208] in the [BxCxHxW] format, where:
B
- batch sizeC
- channelH
- heightW
- widthChannel order is RGB
.
One image of the shape [1x3x208x208] in the [BxCxHxW] format, where:
B
- batch sizeC
- channelH
- heightW
- widthChannel order is BGR
.
The net outputs a vector descriptor, which can be compared with other descriptors using the cosine distance.
Blob of the shape [1, 512] in the [BxC] format, where:
B
- batch sizeC
- predicted descriptor sizeBlob of the shape [1, 512] in the [BxC] format, where:
B
- batch sizeC
- predicted descriptor sizeYou can download models and if necessary convert them into Inference Engine format using the Model Downloader and other automation tools as shown in the examples below.
An example of using the Model Downloader:
An example of using the Model Converter:
[*] Other names and brands may be claimed as the property of others.
The original model is distributed under the MIT License.