Learn Deep Learning for Object Detection!
Budget: $30 – $250 USD
I will pay a $35 milestone for each completed notebook (described below), so please do not bid higher than $35.
Modify 1 (or more) of the google colab (cloud GPU) notebooks here: https://models.roboflow.com/object-detection to train and test on this (mitosis counting) dataset:
https://drive.google.com/drive/folders/1ST0rbcbQAcicYD_nltFsYQMhQOYaNjCW?usp=sharing
It may either be YOLOX, YOLOv5, YOLOR, Scaled-YOLOv4, or EfficientDet-D0-D7.
Please, let me know which one you are starting with.
The test and train annotations are in the folder linked above with the images and are labeled test.txt and train.txt (image x1 y1 x2 y2). There is ONE type of object, with only ONE size bounding box labeled throughout the 2 different image sizes.
Show the mAP for each epoch up to 100.
If you do extra data augmentation for increased mAP and include the code/instructions, I will pay you bonus! but, be sure to include all code either way, including any data preprocessing (but that should not be necessary).
Note: I am a former ML instructor and I gave this project to my students for extra credit. Successful implementations achieved a mAP of 0.7 to 0.9 with data augmentation.
Modify 1 (or more) of the google colab (cloud GPU) notebooks here: https://models.roboflow.com/object-detection to train and test on this (mitosis counting) dataset:
https://drive.google.com/drive/folders/1ST0rbcbQAcicYD_nltFsYQMhQOYaNjCW?usp=sharing
It may either be YOLOX, YOLOv5, YOLOR, Scaled-YOLOv4, or EfficientDet-D0-D7.
Please, let me know which one you are starting with.
The test and train annotations are in the folder linked above with the images and are labeled test.txt and train.txt (image x1 y1 x2 y2). There is ONE type of object, with only ONE size bounding box labeled throughout the 2 different image sizes.
Show the mAP for each epoch up to 100.
If you do extra data augmentation for increased mAP and include the code/instructions, I will pay you bonus! but, be sure to include all code either way, including any data preprocessing (but that should not be necessary).
Note: I am a former ML instructor and I gave this project to my students for extra credit. Successful implementations achieved a mAP of 0.7 to 0.9 with data augmentation.