This repo contains semantic segmentation models for Stanford Drone Dataset and for Semantic Drone Dataset

Detailed solution including visualization results and links to model weights you can find in solution.md.

Stanford Drone Dataset https://cvgl.stanford.edu/projects/uav_data/
Semantic Drone Dataset https://www.tugraz.at/index.php?id=22387

Stanford Drone Dataset

1. download dataset;
2. unpack into data/stanford_drone folder, it should contain annotations/ and videos/ subfolders;
3. run python unsupervised_methods/vanilla_background.py to prepare segmentation masks based on background subtraction;
4. each video folder now should contain box_masks/, frames/, seg_masks/ subfolders and boxes.csv;
5. run python supervised_models/train_stanford.py for training;
6. run tensorboard --logdir=lightning_logs/version_0 to see logs;
7. run python supervised_models/inference.py for inference on validation set. Don't forget to point appropriate checkpoint inside inference.py;

Semantic Drone Dataset

1. download dataset (https://www.kaggle.com/bulentsiyah/semantic-drone-dataset);
2. unpack into data/SDD folder, it should contain RGB_color_image_masks/, semantic_drone_dataset/ subfolders and class_dict_seg.csv;
3. run python supervised_models/sdd_dataset.py to prepare 1-channel (not colored) segmentation masks, categories encoded as int labels: {0:"default", 1:"car", 2:"person", 3:"bicycle"};
4. data/SDD now should contain value_masks subfolder with value masks for each dataset image;
5. run python supervised_models/train_sdd.py for training;
6. run tensorboard --logdir=lightning_logs/version_0 to see logs;
7. run python supervised_models/inference.py for inference on validation set. Don't forget to point appropriate checkpoint and model inside inference.py;