This is an implementation of Mask RCNN on a dataset using the code from https://github.com/matterport/Mask_RCNN
100 Images collected from google to detect 3 objects - O (any circle or oval ), F letter and Star ★.
the dataset is under dataset folder with subfolders train (90 images) and val (10 images)
python3 './ofstar.py' train --dataset='./dataset/' --weights=coco (to train using pretrained coco model) or, python3 './ofstar.py' train --dataset='./dataset/' --weights=last (to start train model form last saved model)
#saved model will be found in ./logs folder created while start training.
you could downlaod from this link below and save to your present working directory. https://drive.google.com/drive/folders/1hW0V-X3W083hcPlCodAcS4uyjskgAqh-?usp=sharing
inspect_ofstar_data.ipynb notebook is provided for visulation of tthe dataset. inspect_ofstar_model.ipynb to use the trained model to detect objects of validation dataset.
| On Train | O | F | Star ★ | Total |
|---|---|---|---|---|
| Ground Truth | 942 | 160 | 177 | 1279 |
| Predictions | 678 | 110 | 157 | 945 |
| True Positive | 624 | 108 | 153 | 885 |
| Recall | 0.6624 | 0.6750 | 0.8644 | 0.6919 |
| precision | 0.9204 | 0.9818 | 0.9745 | 0.9365 |
mAP @ IoU=50: 76.02%, Recall @ IoU=50: 69.19%, Precision @ IoU=50 : 93.65%
| On Validation | O | F | Star ★ | Total |
|---|---|---|---|---|
| Ground Truth | 86 | 6 | 20 | 112 |
| Predictions | 54 | 4 | 11 | 69 |
| True Positive | 38 | 3 | 11 | 52 |
| Recall | 0.44 | 0.50 | 0.55 | 0.4643 |
| precision | 0.70 | 0.75 | 1.00 | 0.7536 |
mAP @ IoU= 50: 45.60%, Recall @ IoU=50: 46.43%, Precision @ IoU=50 : 75.36%,
using Mask R-CNN without Masks (Only Bounding Boxes), Results folder has a 3 samples from validation dataset.
