To Reconstruct Quilt:
-
Download videos using video-ids in the data csv files to the directory
${BASE_DIR}/videos/...with subfolders as described below. -
Download ensemble of histopathology classifiers:
-histo_cyto
-all_inclusive
-dino_checkpoint
to the directory ${BASE_DIR}/histo_models/... as described below.
- To generate images and pair them with text run.
BASE_DIR="/path/to/project" python -m main --base_dir ${BASE_DIR}
The data structure is like this:
dataset
├── /path/to/project
│ ├── video
│ │ ├── channel_id
│ │ │ ├── video_id_name
│ │ │ ├──video_id_name.mp4
│ │ │ ...
│ │ │ ├── video_id_name
│ │ │ ├──video_id_name.mp4
│ ├── frames
│ │ ├── channel_id
│ │ │ ├── video_id_name
│ │ │ ├──video_id_name.mp4
│ │ │ ...
│ │ │ ├── video_id_name
│ │ │ ├──video_id_name.mp4
│ ├── histo_models
│ │ ├── cyto_histo.pth.tar
│ │ ├── all_inclusive.pth.tar
│ │ ├── dino_checkpoint.pth.tar
│ ├── quilt_recon.csv
│ ├── quilt_data.csv
- Please note, there may be some alignment issues as differences in stack and mostly videos or frames extracted can effect image alignment.