TensorFlow with compiled wheel & libraries for Docker Home Assistant on Synology DS918+
Works with Docker Home Assistant 0.96.5 (Python 3.7.4).
Wheel Specs:
- TensorFlow 1.14.1
- Python 3.7
- No AVX instruction set
- Only CPU
- Build on DSM 6.2.2-24922 Update 2, INTEL Celeron J3455
Download
- Wheel: https://drive.google.com/file/d/1WmU_NGkJ4rGURgqBtQw7y5NNSVD_sxOA/view?usp=sharing
- Recommended model: http://download.tensorflow.org/models/object_detection/faster_rcnn_inception_v2_coco_2018_01_28.tar.gz
- Already compiled libraries: https://drive.google.com/file/d/1jQRqsSVi2hr9G6hWGwjgyfvLMOVTHHGE/view?usp=sharing
Install:
1. Download object_detection.zip, unzip & copy to:
/HA config dir/tensorflow/object_detection/
2. Download recommended model & copy to:
/HA config dir/tensorflow/faster_rcnn_inception_v2_coco_2018_01_28/
3. Download wheel & copy to:
/HA config dir/tensorflow/
4. Create tmp folder:
/HA config dir/tensorflow/tmp/
5. Edit HA config:
image_processing:
- platform: tensorflow
scan_interval: 600
source:
- entity_id: camera.*
- entity_id: camera.*
- entity_id: camera.*
- entity_id: camera.*
file_out:
- "/config/tensorflow/tmp/{{ camera_entity.split('.')[1] }}_latest.jpg"
- "/config/tensorflow/tmp/{{ camera_entity.split('.')[1] }}_{{ now().strftime('%Y%m%d_%H%M%S') }}.jpg"
model:
graph: /config/tensorflow/faster_rcnn_inception_v2_coco_2018_01_28/frozen_inference_graph.pb
categories:
- cat
- person
- backpack
- umbrella
- handbag
- suitcase
- bottle
- wine glass
- cup
- fork
- spoon
- knife
- banana
- apple
- sandwich
- orange
- broccoli
- carrot
- pizza
- donut
- cake
- chair
- potted plant
- dining table
- toilet
- laptop
- mouse
- remote
- keyboard
- cell phone
- microwave
- oven
- book
- scissors
- hair drier
- toothbrush
5. Install wheel inside HA docker container via terminal:
docker exec -it homeassistant /bin/bash
pip3 install /config/tensorflow/tensorflow-1.14.1-cp37-cp37m-linux_x86_64.whl
6. Reboot HA