watermark-detector

Instructions

Training

1. Download resnet50 backbone from Memegle google drive at Memegle/database/watermark-detector-related-data/retinanet_resnet50_fpn.pth. Put this backbone in ./retinaNet/backbone/.

2. Download watermark logos from Memegle google drive at Memegle/databse/watermark-detector-related-data/watermark-logos/. Put them in ./watermark_generator/watermarks.

3. Install related packages as per missing requirements (TODO: generate a requirements.txt).

4. Run the following code: cd scripts, then bash ./train_from_scratch.sh.

Predicting

1. Put testing images under ./data/test_images/.

2. Run the following code: cd retinaNet, then python3 predict.py.

3. See the predictions in ./data/predictions.txt.