"How about checking your data before going deeper?"
Use TFRecord Viewer to browse contents of TFRecord files with object detection/classification annotations.
This viewer is wrapper around TFRecord Viewer project by Milan Sulc TFRecord-Viewer.
You can upload (upto 1000) TFRecord files, and check them using a carousel view, you can select multiple files, and either get list of selected files or download the selected ones in a zip file. Be warned that although app allows you to upload 100o tfrecords file, if your computer's memory won't be enough, it would crash.
You can run this project standalone or using Docker. Using Docker is recommended, make sure to have Docker and docker-compose installed, then run:
docker-compose up
Then in your browser (tested with Chrome, Microsoft Edge), navigate to:
http://localhost:3000
If you want to run without Docker, make sure you have Python3 and Tensorflow 1.15 installed. Tensorflow is not listed as dependency in requirements.txt, as base Docker image (tensorflow/tensorflow:1.15.0-py3) has already Tensorflow installed. Then install all other dependencies with:
npm install
pip3 install -r requirements.txt
After that you can run:
npm run runall
Important note: Make sure ports 3000, 5000, and 8000 are not in use before running this project.
On the upload screen, if you select mixed files, only files with tfrecord extension will be uploaded, other file types will be ignored.
On the gallery view screen, you can navigate with right-arrow, left-arrow keys on the keyboard, and select the shown file with space key.
On the summary page, you can concatenate the list of files by line-break, or comma, then you can copy to clipboard. Or you can download the selected files in a zip file.