Please read the Site Policy of GitHub and understand the usage conditions.
This repository provides sample AI models and inference scripts for Raspberry Pi AI Camera Module.
Before running the Python demos, please make sure to install OpenCV:
sudo apt install python3-opencv python3-munkres
Python demo is in models folder.
The model (*.rpk), label, and config files are not included in this repository. These are output files from Brain Builder.
In anomaly-detection/brainbuilder
python app.py --model YOUR_MODEL.rpk --config neurala_hifi_ppl_parameters.json --fps 15
In classification/brainbuilder
python app.py --model YOUR_MODEL.rpk --labels labels.txt --fps 15
In object-detection/brainbuilder
python app.py --model YOUR_MODEL.rpk --labels labels.txt --fps 15
The model (*.rpk), label are not included in this repository. These are output files from aitrios-rpi-training-samples
In anomaly-detection/rd4ad
python app.py --model YOUR_MODEL.rpk --fps 15 --overlay
In classification/mobilenet_v2
In object-detection/efficientdet_lite0_pp
In segmentation/deeplabv3plus
In pose_estimation/higherhrnet_coco
In the table below you can find information regarding all models in the model zoo.
Please use the following models or datasets in accordance with the licenses of their respective sources.
| Task |
Model |
Dataset |
Float model accuracy |
Quantized model accuracy |
Input resolution |
Original model repository |
License |
URL for Images/Annotation |
| Classification |
efficientnet_lite0 |
Imagenet_1k |
0.753 |
0.753 |
224x224 |
trimm repository |
models/classification/ efficientnet_lite0/LICENSE.md |
https://www.image-net.org/ |
| Classification |
efficientnetv2_b0 |
Imagenet_1k |
0.764 |
0.767 |
224x224 |
Keras Applications |
models/classification/ efficientnetv2_b0/LICENSE.md |
https://www.image-net.org/ |
| Classification |
efficientnetv2_b1 |
Imagenet_1k |
0.769 |
0.777 |
240x240 |
Keras Applications |
models/classification/ efficientnetv2_b1/LICENSE.md |
https://www.image-net.org/ |
| Classification |
efficientnetv2_b2 |
Imagenet_1k |
0.779 |
0.770 |
260x260 |
Keras Applications |
models/classification/ efficientnetv2_b2/LICENSE.md |
https://www.image-net.org/ |
| Classification |
efficientnet_bo |
Imagenet_1k |
0.739 |
0.721 |
224x224 |
Torchvision |
models/classification/ efficientnet_bo/LICENSE.md |
https://www.image-net.org/ |
| Classification |
resnet18 |
Imagenet_1k |
0.685 |
0.686 |
224x224 |
trimm repository |
models/classification/ resnet18/LICENSE.md |
https://www.image-net.org/ |
| Classification |
regnety_002 |
Imagenet_1k |
0.696 |
0.694 |
224x224 |
trimm repository |
models/classification/ regnety_002/LICENSE.md |
https://www.image-net.org/ |
| Classification |
regnetx_002 |
Imagenet_1k |
0.682 |
0.684 |
224x224 |
trimm repository |
models/classification/ regnetx_002/LICENSE.md |
https://www.image-net.org/ |
| Classification |
regnety_004 |
Imagenet_1k |
0.734 |
0.738 |
224x224 |
trimm repository |
models/classification/ regnety_004/LICENSE.md |
https://www.image-net.org/ |
| Classification |
mobilevit_xs |
Imagenet_1k |
0.724 |
0.723 |
256x256 |
trimm repository |
models/classification/ mobilevit_xs/LICENSE.md |
https://www.image-net.org/ |
| Classification |
mobilevit_xxs |
Imagenet_1k |
0.674 |
0.674 |
256x256 |
trimm repository |
models/classification/ mobilevit_xxs/LICENSE.md |
https://www.image-net.org/ |
| Classification |
levit_128s |
Imagenet_1k |
0.583 |
0.623 |
224x224 |
Facebook research |
models/classification/ levit_128s/LICENSE.md |
https://www.image-net.org/ |
| Classification |
mobilenet_v2 |
Imagenet_1k |
0.74 |
0.726 |
224x224 |
Keras Applications |
models/classification/ mobilenet_v2/LICENSE.md |
https://www.image-net.org/ |
| Classification |
shufflenet_v2_x1_5 |
Imagenet_1k |
0.725 |
0.722 |
224x224 |
trimm repository |
models/classification/ shufflenet_v2_x1_5/LICENSE.md |
https://www.image-net.org/ |
| Classification |
mnasnet1.0 |
Imagenet_1k |
0.731 |
0.732 |
224x224 |
trimm repository |
models/classification/ mnasnet1.0/LICENSE.md |
https://www.image-net.org/ |
| Classification |
squeezenet1.0 |
Imagenet_1k |
0.576 |
0.576 |
224x224 |
trimm repository |
models/classification/ squeezenet1.0/LICENSE.md |
https://www.image-net.org/ |
| Object Detection |
efficientdet_lite0 |
COCO (mAP) |
0.273 |
0.274 |
320x320 |
efficientdet-pytorch |
models/object-detection/ efficientdet_lite0/LICENSE.md |
http://images.cocodataset.org/zips/train2017.zip (19GB, 118k images) http://images.cocodataset.org/zips/val2017.zip (1GB, 5k images) http://images.cocodataset.org/annotations/annotations_trainval2017.zip |
| Object Detection |
nanodet_plus_416x416_pp |
COCO (mAP) |
0.323 |
0.32 |
416x416 |
Nanodet |
models/object-detection/ nanodet_plus_416x416_pp/LICENSE.md |
http://images.cocodataset.org/zips/train2017.zip (19GB, 118k images) http://images.cocodataset.org/zips/val2017.zip (1GB, 5k images) http://images.cocodataset.org/annotations/annotations_trainval2017.zip |
| Object Detection |
nanodet_plus_416x416 |
COCO (mAP) |
0.332 |
0.332 |
416x416 |
Nanodet |
models/object-detection/ nanodet_plus_416x416/LICENSE.md |
http://images.cocodataset.org/zips/train2017.zip (19GB, 118k images) http://images.cocodataset.org/zips/val2017.zip (1GB, 5k images) http://images.cocodataset.org/annotations/annotations_trainval2017.zip |
| Object Detection |
efficientdet_lite0_pp |
COCO (mAP) |
0.25 |
0.252 |
320x320 |
efficientdet-pytorch |
models/object-detection/ efficientdet_lite0_pp/LICENSE.md |
http://images.cocodataset.org/zips/train2017.zip (19GB, 118k images) http://images.cocodataset.org/zips/val2017.zip (1GB, 5k images) http://images.cocodataset.org/annotations/annotations_trainval2017.zip |
| Pose Estimation |
higherhrnet (coco) - multi person |
COCO |
0.187 |
0.188 |
288x384 |
HRnet |
models/pose-estimation/ higherhrnet_coco/LICENSE.md |
http://images.cocodataset.org/zips/train2017.zip (19GB, 118k images) http://images.cocodataset.org/zips/val2017.zip (1GB, 5k images) http://images.cocodataset.org/annotations/annotations_trainval2017.zip |
| Semantic Segmentation |
Deeplabv3plus |
PASCAL VOC (mIOU) |
0.72 |
0.721 |
320x320 |
Tensorflow |
models/segmentation/ Deeplabv3plus/LICENSE.md |
http://host.robots.ox.ac.uk/pascal/VOC/ |