# Internship
> State-of-the-art deep learning models demo stand
2020 June
Anastasia BS17-RO

## Agenda
* 8.06-14.06: Define goals and objectives. Preliminary plan (may be changed)
* 15.06-21.06
* 22.06-28.06
* 29.06-5.07

#### Goal
Apply Machine Learning Model Zoo on given data to implement demo stand with new computer vision solutions

#### Steps
* Find models, relevant to the data in OpenVINO pre-trained Zoo
* Create repository
* Learn OpenVINO
* Apply models on camera data
* Compare models (OpenVINO workbench)

#### Plan
* 8.06-14.06: Make gitlab repository to store results and report meeting's info and conclusions. Start learning OpenVINO (watch videos, tutorials, read documentation). Take data from camera and start connecting it to OpenVINO models. + find sota models + resnet/vgg/densnet/mobilnet/unet/mask-rcnn
* 15.06-21.06: Try to adapt chosen models for OpenVINO and connect them to data from camera. Compare models, choose assessment criteria + learn OpenVINO workbench.
* 22.06-28.06: Try to launch sota models and convert to openVino
* 29.06-5.07: Add to repository all documentations, finalize scripts, etc to make it easy to use Model Zoo for demo stand.

### 8.06-14.06
Need to use models, which are a part of OpenVino, also use newly invented models(2019-2020).
Links with description of models, which will be used:
1. https://habr.com/ru/company/ods/blog/502786/#4-tracking-objects-as-points
2. https://habr.com/ru/company/ods/blog/505040/

#### Models, chosen

##### Detection:
* People, Vehicles & Bikes Detection (MobileNetV2 + SSD-based network) - https://docs.openvinotoolkit.org/2020.1/_models_intel_person_vehicle_bike_detection_crossroad_1016_description_person_vehicle_bike_detection_crossroad_1016.html
* YOLO v4 - https://arxiv.org/pdf/2004.10934v1.pdf

##### Segmantation:
Semantic segmantation, which assigns 1 of 20 classes to each pixel (for roads) - https://docs.openvinotoolkit.org/latest/_models_intel_semantic_segmentation_adas_0001_description_semantic_segmentation_adas_0001.html
EfficientPS: Efficient Panoptic Segmentation https://arxiv.org/pdf/2004.02307v2.pdf

##### Pose estimation:
There is only one OpenVino pose estimetion model, thus, we both, can try to run the same model - https://docs.openvinotoolkit.org/latest/_models_intel_human_pose_estimation_0001_description_human_pose_estimation_0001.html

Human Pose Estimation with Spatial Contextual Information https://arxiv.org/pdf/1901.01760v1.pdf

#### Gitlab, youtrack accounts
aasya9999