SkalskiP
export csv format can match for Google's Automl?
Hi, I am trying to use make-sense for my own object detection project. Yet, when I want to upload the csv to Google Cloud. It seems like not really supporting. Just wondering if a new feature can add to export format fits GCP/ AWS? Thanks
Hi @herruli, Of course, you can count on importing our CSVs into GCP/ AWS. I suppose this is a small change. Do you have any details of the error? Do you know what caused the error during import?
As https://cloud.google.com/vision/automl/object-detection/docs/quickstart-ui#preparing_a_dataset stated. It's bounding box is in between 0 and 1. Where make-sense used pixel location to label. Also, Google has a four point annotation maybe it's also a nice feature to add?
Thank you for sending me the link to the documentation. As I expected, the changes that need to be added to ensure the ability to import data created in MakeSense to Google Automl seem simple. I don't want to promise anything, but maybe I can write it in a few days. As far as this form of labeling is concerned, we already have an issue on the subject: https://github.com/SkalskiP/make-sense/issues/12#issuecomment-577406093 Please add your opinion on that there. It matters.
Also really need this for a project - trying to load the data using VOC XML format but that's having issues as well - csv seems to be optimal for AutoML/google tutorials on such things:
https://colab.research.google.com/github/google-coral/tutorials/blob/master/retrain_efficientdet_model_maker_tf2.ipynb
Hello 👋 , @herruli, @Sounder317 and @ergoego! I am very sorry for the late reply. Work factors did not give me time to develop MS. I am pleased to see quite a lot of interest in this issue. This is quite an old issue. I must apologise to you for this.
However, I am starting work on the make-sense beta, and I want this issue to be part of it. I'll let you know about the (hopefully) rapid progress.