AI: Object Detection

Evolving documentation of the status quo.

git repo: https://git.picalike.corpex-kunden.de/incubator/ai-experiments/-/tree/master/object_detect

grapex: /home/picalike/tdog_sandbox/object_detect

What we currently use for the fashion detection:

https://github.com/svip-lab/HRNet-for-Fashion-Landmark-Estimation.PyTorch
pretrained network grapex: /home/picalike/tdog_sandbox/object_detect/fasion/pose_hrnet.pth

the network does not output bounding boxes, but fashion landmarks for the dominating product in an image.

Links:

https://github.com/switchablenorms/DeepFashion2 [data set, not publicly accessible]
http://mmlab.ie.cuhk.edu.hk/projects/DeepFashion.html [data set, outdated, not free]
https://github.com/eBay/modanet [model + data only non-commercial use]
https://github.com/hrsma2i/dataset-CFPD [dataset, not free]
https://openaccess.thecvf.com/content_ICCVW_2019/papers/CVFAD/Sidnev_DeepMark_One-Shot_Clothing_Detection_ICCVW_2019_paper.pdf [paper]

To be evaluated, but only bounding persons for persons [COCO classes]:
https://github.com/HRNet/HRNet-Object-Detection

http://sandy.picalike.corpex-kunden.de:8033/docs

https://pytorch.org/hub/nvidia_deeplearningexamples_ssd/

The SSD network is fast and small, but also only predicts bounding boxes for persons [COCO classes]

https://fashionpedia.github.io/home/Fashionpedia_download.html
https://fashionpedia.github.io/home/Model_and_API.html
https://github.com/tensorflow/tpu/tree/master/models/official/detection/projects/fashionpedia#checkpoint
Keywords: deepfashion dataset