This is an action recognition composite model for the driver monitoring use case, consisting of encoder and decoder parts. The encoder model uses Video Transformer approach with MobileNetV2 encoder. It is able to recognize actions such as drinking, doing hair or making up, operating the radio, reaching behind, safe driving, talking on the phone, texting. The full list of recognized actions is located at <omz_dir>/demos/action_recognition_demo/python/driver_actions.txt.
| Metric | Value |
|---|---|
| Source framework | PyTorch* |
The driver-action-recognition-adas-0002-encoder model accepts video frame and produces embedding. Video frames should be sampled to cover ~1 second fragment (i.e. skip every second frame in 30 fps video).
| Metric | Value |
|---|---|
| GFlops | 0.676 |
| MParams | 2.863 |
Image, name: 0, shape: 1, 3, 224, 224 in the format B, C, H, W, where:
B- batch sizeC- number of channelsH- image heightW- image width
Expected color order is BGR.
The model outputs a tensor with the shape 1, 512, 1, 1, representing embedding of processed frame.
The driver-action-recognition-adas-0002-decoder model accepts stack of frame embeddings, computed by driver-action-recognition-adas-0002-encoder, and produces prediction on input video.
| Metric | Value |
|---|---|
| GFlops | 0.147 |
| MParams | 4.205 |
An embedding image, name: 0, shape: 1, 16, 512 in the format B, T, C, where:
B- batch sizeT- duration of input clipC- dimension of embedding
The model outputs a tensor with the shape 1, 9, each row is a logits vector of performed actions.
The model can be used in the following demos provided by the Open Model Zoo to show its capabilities:
[*] Other names and brands may be claimed as the property of others.