Skip to content

Latest commit

 

History

History
68 lines (35 loc) · 2.51 KB

README_SIGNET_RING_DETECTION.md

File metadata and controls

68 lines (35 loc) · 2.51 KB
Raw

Installation

  1. Follow the installation instructions from README.md

  2. (Optional step if not using the docker container) Make sure to also install the requirements of this project (we do recommend using a virtual environment)

    pip install -r requirements.txt

  3. Place your signet ring dataset on a suitable folder which can be accessed from this application.

  4. Make a copy of the settings template.

    cp settings_template.py settings.py

  5. Open your settings.py and modify the paths for your database properly and train/test pickle files properly (the latter will be created in the next step).

  6. Create split your database and create your pickle files

    from utils.data import get_or_create_train_test_files

    get_or_create_train_test_files(test_size=0.8, random_state=42, shuffle=True, force_create=True)

  7. The configuration files for training and testing are located at:

    config/yolov3_default_digestpath.cfg

    config/yolov3_eval_digestpath.cfg

  8. If you want to re-calculate the anchor boxes for your dataset you can do it by running the function recalculate_anchor_boxes_kmeans_iou, then you just need to update the ANCHORS on both configuration files.

    from constants import Dataset

    from utils.utils import recalculate_anchor_boxes_kmeans_iou

    new_anchors = recalculate_anchor_boxes_kmeans_iou(Dataset.SIGNET_RING, print_results=True, num_centroids=9)

    print(new_anchors)

  9. Based on the size of your images and hardware especifications you should update the following variables from the configuration files: MAXITER, BATCHSIZE, SUBDIVISION, IMGSIZE.

  10. Other customizations could be done on CONFTHRE and NMSTHRE.

Training

import os

os.system('python train.py --weights_path weights/darknet53.conv.74 --tfboard True --checkpoint_interval=50 --eval_interval=50')

Evaluation

Quick example using our provided checkpoint

  1. Get sample checkpoint. Download it at the same folder were input_sample is located.

    git clone [email protected]:giussepi/PyTorch_YOLOv3_sample_checkpoint.git

  2. Rename PyTorch_YOLOv3_sample_checkpoint folder.

    mv PyTorch_YOLOv3_sample_checkpoint checkpoints

  3. Run the evaluation command.

    import os

    os.system('python train.py --cfg config/yolov3_eval_digestpath.cfg --eval_interval 1 --checkpoint "checkpoints/confthre_0_dot_8/snapshot17350.ckpt"')