One Piece characters detector

An object detector trained with a Kaggle GPU on One Piece images, using Tensorflow and a fine-tuned ssd resnet50.

**Deployed on my personal Docker Hub repository

**Kaggle Notebook link: Kaggle notebook

**One Piece Detector SavedModel : Drive

Model and configuration summary

• Class count : 1
• Labels : ["character"]
• Model type : detection
• Base model : ssd_resnet50_v1_fpn_640x640_coco17_tpu-8

Results : screenshots

Sample video

Test with 4 characters

Useful links

How to run inferences on a video : commnand line runner

Note : Before running the script, you must first download the One Piece object detector model here.

Args

Execution example

Check the sample command for executing the command line runner

Tensorflow Serving container test script

• On GitHub
• Label map

Training notebook

• Kaggle
• GitHub

How to concatenate mutliple TFRecord files into the Training and Validation sets

• Kaggle
• GitHub

Training pipeline configuration file : Here

Prerequisites

• Python 3.7 or higher
• Tensorflow 2.6 or higher
• Tensorflow Object Detection API
• NumPy
• Matplotlib