Overwatch-AI-Aimbot

Object Detection with YOLO/MOBILENET architecture

2 Pretrained Models are included under Workspace/pre-trained-models:

1. SSD_MobileNet (320x320)
2. SSD_MobileNet (640x640)

Unzip these .tar.gz files using unzip.py

For more pretrained models, download at https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_detection_zoo.md

Required Packages

1. Tensorflow Object Detection API:

   https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/install.html

2. LabelImg Tool: https://github.com/tzutalin/labelImg

3. Protocol Buffers: https://github.com/protocolbuffers/protobuf/releases

4. Pretrained Models: https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/tf2_detection_zoo.md

File Structure

─ Overwatch-AI-Aimbot/
    ├─ Workspace/
      ├─ annotations/
      ├─ images/
      	├─ screenshots/
      	├─ train/
      	├─ test/
      ├─ my_models/
      ├─ pre-trained-models/
    ├─ processImg.py
    ├─ unzip.py
    ├─ testDetection.py
    ├─ train.py
    ├─ generate_tfrecord.py
    
─ RealTimeObjectDetection/
    ├─ Tensorflow
      ├─ models/
      	├─ research/
      		├─ object_detection/
      ├─ scripts/
      	├─ generate_tfrecord.py
    ├─ labelImg/
      ├─ labelImg.py

Overwatch-AI-Aimbot:

• Clone from this repository
• Place all images in Workspace/images/
• Place all pretrained-models in Workspace/pre-trained-models

RealTimeObjectDetection:

• Create /RealTimeObjectDetection folder in same directory as /Overwatch-AI-Aimbot folder
• /Tensorflow folder contains the "models" repository, which is cloned via https://github.com/tensorflow/models
• /Tensorflow folder contains a "/scripts" folder. Create this folder and paste the generate_tfrecord.py file here
• /RealTimeObjectDetection contains the "labelImg" repository, which is cloned via https://github.com/tzutalin/labelImg

preprocessImg.py

Used to rename image filenames in ascending order (E.g 1.png, 2.png, etc.)

You must rename the images first, before doing annotations using labelImg.py

python preprocessImg.py [--filepath]

--filepath: Folder containing images

E.g: python processImg.py Workspace/images/screenshots

unzip.py

Used to unzip .tar.gz or .tar files. Useful for unzipping pretrained model files

python unzip.py

train.py

Used to process data train model

• Generated Tfrecords & labelmap will be in Workspace/annotataions
• Model weights will be saved to Workspace/my_model/{{CUSTOM_MODEL_NAME}}/

Will automatically

• Split images into train/test folders
• Edit pipeline.config file
• Create tfrecords from images

python train.py [-L --labels] [-E --epochs] [-N --name] [-T --split] [-s --trainSize] [-M --model]

--labels: (str) Labels in json format. Must follow {'name': ..., 'id': ...} convention, where 'name' is the label given in labelImg.py tool.
Separate labels using triple asterix (E.g "{...}* * *{...}" )

--epochs: (int) Number of training epochs to run

--name: (str) Name of custom model

--split: (str; optional) "yes" or "no" (Defaults to 'no'). If "yes", will automatically split images & annotations in /screenshots folder, to /train and /test folder using trainSize

--trainSize: (float; optional) Training size (0-1, defaults to 1)

--model: (str; optional) Pretrained model to use. Specify folder name containing pretrained model (located in Workspace/pre-trained-models/)

(Defaults to ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8)

E.g: python train.py -L "{'name': 'head', 'id': 1 }***{'name': 'body', 'id': 2 }" -E 50 -N "overwatch_mobnet" -T yes -S 0.7 -M "centernet_resnet50"

Internally, this will create:

• label=[{'name': 'head', 'id': 1 }, {'name': 'body', 'id': 2 }]
• Run for 50 epochs
• Create a model named "Overwatch_mobnet"
• Split the files in /screenshot folder, with 70% going to /train and the remaining 30% going to /test
• Train using the "centernet_resnet50" pretrained model, which can be found at Workspace/pre-trained-models/centernet_resnet50
• Run training for the model, where outputs can be found in my_models/overwatch_mobnet/

testDetection.py

Used to test custom_model

python testDetection.py

generate_tfrecord.py

Used for creation of TFRecord format from image files

Paste generate_tfrecord.py file to /RealTimeObjectDetection/Tensorflow/scripts