Converting the .csv to .xml

[m-kashani]

• [ ] https://github.com/howl0893/custom-object-detection-datasets/blob/master/python/coco_json_utils.py

• [ ] https://github.com/akTwelve/cocosynth/blob/master/docs/getting-started.md

  • [ ] https://github.com/howl0893/custom-object-detection-datasets

  • [ ] https://www.youtube.com/watch?v=jftZBfMZj8k&list=PLy-L-4xQRIuNm4tN9RIufCC6sqjuw1mAj&index=1

  • [ ] https://github.com/akTwelve/cocosynth/tree/cd9d583f6c4ce2c0fc185d8dedb4788b5b7af0f0

---

To Do Next:

• [ ] get it done after the While(evaluation, preprocessing ) + Comparison of Detectron2 output layer. -> I hardly couldn't. Very boring and not ... ( Not now.)

• [ ] Alireza: https://patrickwasp.com/create-your-own-coco-style-dataset/ So -> It's talking about the binary mask that I do not have yet, but it says that it's very easy to create.

• https://www.kaggle.com/julian3833/2-understanding-and-plotting-rle-bounding-boxes -> The new issue is creating the rle from BBOX coordination. (Later.)

• [ ] -> What about this? https://stackoverflow.com/questions/50368683/set-white-color-outside-boundingbox-python-opencv

• [ ] https://discuss.pytorch.org/t/coco-dataset-from-custom-semantic-segmentation-dataset-for-detectron2/72266

download, decompress the data

!wget https://github.com/matterport/Mask_RCNN/releases/download/v2.1/balloon_dataset.zip
!unzip balloon_dataset.zip > /dev/null

• [ ] -> Checking out the balloon data-set again. + Annotation.

• [ ] -> https://detectron2.readthedocs.io/modules/structures.html#detectron2.structures.BoxMode
• [ ] -> going through the detectron issues349 :+1:
• [x] Installing the : https://github.com/matterport/Mask_RCNN/issues/6
• [x] https://github.com/facebookresearch/detectron2/issues/349 (Yess)

---

ASK if you need to email or install the software yourself.

Sol 1. Email and make sure

• [ ]

---

Sol 2. Shortcut?

• [ ] -install: https://github.com/microsoft/VoTT

Sol 3. PeerCode or out source

• [ ] Peer Coding + Reverse engineering: https://github.com/Shenggan/BCCD_Dataset/blob/master/BCCD/Annotations/BloodImage_00000.xml
• [ ] https://github.com/Shenggan/BCCD_Dataset/blob/master/export.py (reverse engineering)

• [ ] - https://detectron2.readthedocs.io/modules/structures.html#detectron2.structures.BoxMode

- [ ] - Code6.py is under construction.

---

Just generate a new code and get it done.

• [ ] Prepare for coco dataset following this instruction this instruction. https://github.com/facebookresearch/detectron2/tree/master/datasets

• [ ] Avoid Training for now.

  Training

ImageNet Pretrained Models

We provide backbone weights pretrained on ImageNet-1k dataset.

• VoVNetV2-19-Slim-DW
• VoVNetV2-19-Slim
• VoVNetV2-19-DW
• VoVNetV2-19
• VoVNetV2-39
• VoVNetV2-57

• VoVNetV2-99

  • [ ] ?????? To train a model, run:

    python /path/to/`model-detectron2`/train_net.py --config-file /path/to/`my_dataset`/configs/<config.yaml>

COCO Pretrained Models

https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md

• [1]
• [2]
• [3]
• [ ]

• [ ]

  • [ ] - https://github.com/tensorflow/models/blob/master/research/deeplab/g3doc/model_zoo.md

=== Object detection and recognition === https://en.wikipedia.org/w/index.php?title=List_of_datasets_for_machine-learning_research&action=edit&section=4

[m-kashani]

to do -> is crowd can be calculated by the histogram.