Skip to content

models.md

Latest commit

 

History

History
95 lines (63 loc) · 3.39 KB

models.md

File metadata and controls

95 lines (63 loc) · 3.39 KB

Single Mouse Segmentation

Original Training Code: https://github.com/KumarLabJax/MouseTracking Trained Models:

TSF Model

The segmentation model was exported using code that resides in the obj-api codebase. This code was largely based on tensorflow example code for optimizing and freezing a model.

Single Mouse Pose

Original Training Code: https://github.com/KumarLabJax/deep-hrnet-mouse Trained Models:

Pytorch Model

The pytorch model is the released model.

Multi-Mouse Pose

Original Training Code: https://github.com/KumarLabJax/deep-hrnet-mouse Trained Models:

  • Top-down: In Progress
  • Bottom-up: (Not published)

Pytorch Model

The pytorch model is the model saved by the original hrnet code.

Multi-Mouse Segmentation

Original Training Code: fork of https://github.com/google-research/deeplab2 Trained Models:

  • Panoptic Deeplab: Not yet released

TFS Model

deeplab2 provides export_model.py. This transforms the checkpoint into a tensorflow serving model.

python3 /deeplab2/deeplab2/export_model.py --checkpoint_path /deeplab2/trained_model/ckpt-125000 --experiment_option_path /deeplab2/trained_model/resnet50_os16.textproto --output_path tfs-models/multi-mouse-segmentation/panoptic-deeplab/

Static Objects

Arena Corners

Original Training Code: In Progress Trained Models:

  • Object Detection API (2022): Not yet released

TFS Model

Export the model using the tf-obj-api exporter (in obj-api environment):

python /object_detection/models/research/object_detection/exporter_main_v2.py --input_type image_tensor --pipeline_config_path /object_detection/code/tf-obj-api/corner-detection/single-object-testing/pipeline.config --trained_checkpoint_dir /media/bgeuther/Storage/TempStorage/pose-validation/movenet/arena_corner/output_models/ --output_directory /media/bgeuther/Storage/TempStorage/trained-models/static-objects/obj-api-corner/

Note that this needs to be run in the folder with annotations if the config points to label_map.pbtxt locally. /media/bgeuther/Storage/TempStorage/pose-validation/movenet/arena_corner/ is the location of these annotations.

Food Hopper

Original Training Code: In Progress Trained Models:

  • Object Detection API (2022): In Progress

TFS Model

Export the model using the tf-obj-api exporter (in obj-api environment):

python /object_detection/models/research/object_detection/exporter_main_v2.py --input_type image_tensor --pipeline_config_path /object_detection/code/tf-obj-api/food-detection/segmentation/pipeline.config --trained_checkpoint_dir /media/bgeuther/Storage/TempStorage/pose-validation/movenet/food_hopper/output_models/ --output_directory /media/bgeuther/Storage/TempStorage/trained-models/static-objects/obj-api-food/

Note that this needs to be run in the folder with annotations if the config points to label_map.pbtxt locally. /media/bgeuther/Storage/TempStorage/pose-validation/movenet/food_hopper/ is the location of these annotations.

Lixit

Original Training Code: In Progress Trained Models:

  • DeepLabCut: In Progress

Dynamic Objects

Fecal Boli

Original Training Code: https://github.com/KumarLabJax/deep-hrnet-mouse Trained Models:

  • fecal-boli (2020): Not yet published.

Pytorch Model

The pytorch model is the model saved by the original hrnet code.