ODG-Generation

1. Generate datasets

Load models, randomly build scenes, render image data, compute grab labels, save data, data enhancement.

operating steps:

(1) Download obj_models.zip from https://drive.google.com/file/d/1QLAubfa6nCkmRGeCZuy44lFfwKWobxVa/view?usp=drive_link and unzip.

(2) Open scripts\dataset\render_grasp_data.py, modify model_path to path of obj_models, and modify save_path to dataset path.

(3) run python scripts\dataset\render_grasp_data.py.

(4) The dataset is divided into two parts: training set and test set.

(5) Open scripts\dataset\create_dataset.py, modify src_path to path of training set or test set, and modify dst_path for training network.

(6) run python scripts\dataset\create_dataset.py.

2. Visualize mask and bbox labels

(1) Open scripts\dataset\visual_mask_bbox.py, modify dataset_path to dataset path.

(2) run python scripts\dataset\visual_mask_bbox.py

3. Benchmark

(1) Download backbones.zip from https://drive.google.com/file/d/19FaSnMmUCa8yjpkX-jgucA0D5wRsNWe0/view?usp=drive_link and place files in grasp_methods\ckpt.

(2) Open scripts\grasp_experiments\test_acc.py, modify obj_model_path to path of obj_models.

(3) run scripts\grasp_experiments\test_acc.py

(4) Open scripts\grasp_experiments\test_AP.py, modify obj_model_path to path of obj_models.

(5) run scripts\grasp_experiments\test_AP.py

4. Future work

Translate comments into English.