Skip to content

Latest commit

 

History

History
55 lines (44 loc) · 1.3 KB

File metadata and controls

55 lines (44 loc) · 1.3 KB

Landmark dataset

For access to the dataset, please contact Rhodri Davies (rhodri.davies@ucl.ac.uk).

Download Dataset

Note

It is recommended to download the dataset to ~/.cache/cinema_datasets/landmark as the integration tests are based on this path.

The files should have the following structure:

lax_2c.csv
lax_2c/
├── images/
│   ├── 0001__00.png
│   ├── 0001__09.png
│   ├── ...
├── masks/
│   ├── 0001__00.png
│   ├── 0001__09.png
│   ├── ...
lax_4c.csv
lax_4c/
├── images/
│   ├── 0001__00.png
│   ├── 0001__09.png
│   ├── ...
├── masks/
│   ├── 0001__00.png
│   ├── 0001__09.png
│   ├── ...

Training

The following command can be used to train/fine-tune the model on the dataset (please adjust the model.ckpt_path and data.dir accordingly):

# from scratch
landmark_seg_train data.dir=~/.cache/cinema_datasets/landmark/processed
# fine-tune
landmark_seg_train data.dir=~/.cache/cinema_datasets/landmark/processed model.ckpt_path=

Evaluation

The following command can be used to evaluate the model (please adjust the model.ckpt_path accordingly):

landmark_seg_eval --data_dir ~/.cache/cinema_datasets/landmark/processed --ckpt_path