SLAB

Step 1

• Download the TextRecognitionDataGenerator and place the unzipped folder in this repo (inside CNN/). I've added it to my filepath but placed it in the .gitignore as it's not our work.

Step 2

• Move the TextRecognitionDataGenerator/TextRecognitionDataGenerator/dicts/ and TextRecognitionDataGenerator/TextRecognitionDataGenerator/fonts/ folders to the same parent folder as shown above.

Step 3

• Enter CNN/ and run python detection_generation.py [keyword] to generate a detection training set with that specific word. As of now, the training set will be 10,000 images strong. This can be edited via detection_generation.py. The training set will be saved in data_generation/pictures and the corresponding xml files will be stored in data_generation/labels. This data is not tracked by github.

Step 4

1. We are using the Yolo algorithm for the detection of shop signs.
2. Follow this tutorial to install OR cd into ./darkflow and run python3 setup.py build_ext --inplace
3. Download ckpt folder and place it into ./darkflow from here
4. Download build_graph folder and place it into ./darkflow from here
5. Download bin folder and place it into ./darkflow from here

Training with Pre-trained Pascal VOC weights:

The steps below assume we want to use tiny YOLO and our dataset has 3 classes

1. Create a copy of the configuration file tiny-yolo-voc.cfg and rename it according to your preference tiny-yolo-voc-3c.cfg (It is crucial that you leave the original tiny-yolo-voc.cfg file unchanged, see below for explanation).

2. In tiny-yolo-voc-3c.cfg, change classes in the [region] layer (the last layer) to the number of classes you are going to train for. In our case, classes are set to 3.

   ...
   
   [region]
   anchors = 1.08,1.19,  3.42,4.41,  6.63,11.38,  9.42,5.11,  16.62,10.52
   bias_match=1
   classes=3
   coords=4
   num=5
   softmax=1
   
   ...

3. In tiny-yolo-voc-3c.cfg, change filters in the [convolutional] layer (the second to last layer) to num * (classes + 5). In our case, num is 5 and classes are 3 so 5 * (3 + 5) = 40 therefore filters are set to 40.

   ...
   
   [convolutional]
   size=1
   stride=1
   pad=1
   filters=40
   activation=linear
   
   [region]
   anchors = 1.08,1.19,  3.42,4.41,  6.63,11.38,  9.42,5.11,  16.62,10.52
   
   ...

4. Change labels.txt to include the label(s) you want to train on (number of labels should be the same as the number of classes you set in tiny-yolo-voc-3c.cfg file). In our case, labels.txt will contain 3 labels.

   label1
   label2
   label3
   

5. Reference the tiny-yolo-voc-3c.cfg model when you train.

   flow --model cfg/tiny-yolo-voc-3c.cfg --load bin/tiny-yolo-voc.weights --train --annotation train/Annotations --dataset train/Images

• Why should I leave the original tiny-yolo-voc.cfg file unchanged?

  When darkflow sees you are loading tiny-yolo-voc.weights it will look for tiny-yolo-voc.cfg in your cfg/ folder and compare that configuration file to the new one you have set with --model cfg/tiny-yolo-voc-3c.cfg. In this case, every layer will have the same exact number of weights except for the last two, so it will load the weights into all layers up to the last two because they now contain different number of weights.

6. Run ./flow --model ./cfg/tiny-yolo-voc-<CUSTOM INPUT>.cfg --load bin/tiny-yolo-voc.weights --train --dataset <DATASET DIRECTORY> --annotation <ANNOTATIONS DIRECTORY> --gpu 1.0

7. Run python yolo_predict.py <DATA_DIRECTORY> -> This will save the images with the bounding boxes to your desired save_dir or default: ./darkflow/predictions/.