Open pccoe-co opened 2 years ago
Hi, Thanks for reaching out. I have not tested it myself (I have done the conversion the other way around). It looks like several tools such as Roboflow would allow you to do that. Never used it myself so check carefully the condition of usage.
Alternatively, our Detectron2 notebook should be relatively easy to modify to take annotation in a JSON format. But you would still need one JSON file for training and another that contains your validation dataset.
I hope this helps a little
Cheers
Guillaume
@pccoe-co I have looked at both files classifications.csv
and classifications.json
and I notice the fields nucleus_x
and nucleus_y
. I assume these to be the center coordinates of the nucleus you wish to detect and classify. However, the Pascal VOC format uses bounding boxes, not center points, in the format [x_min, y_min, x_max, y_max]
in pixel integers. More info here.
I think that you will need to adjust your raw data to use bounding boxes, or perhaps make an assumption that a bounding box has a constant height and width.
Regards,
Junel S.
@guijacquemet thanks for the tool name. I converted my files in xml (for sake of uploading on github changed extension to txt) 366_png.rf.057ead6f7eeb6306fa67af11d3f7c009.txt
But i am getting error while running cell 3.2 for augmentation. IndexError: child index out of range. Also while training mAP is zero and loss is not changing at all
How to solve this issue?
Hi, Are your annotations read properly in the notebook? are your images/annotations displayed properly when you run cell 3.1?
Cheers Guillaume
Hi, It does read annotation properly.
in config.json file i can see anchors are mostly zero. could this be the reason for zero mAP and infinite loss value?
I have image dataset and its annotation in json and a csv file. how do i convert it into pascal voc xml format so that i can use it for object detection using yoloV2 (https://github.com/HenriquesLab/ZeroCostDL4Mic/blob/master/Colab_notebooks/YOLOv2_ZeroCostDL4Mic.ipynb). my aim is to localize and classify cells in an image.
link to json and csv file (https://drive.google.com/drive/folders/1WJyCZMDaoroa6K4s4AbZQygFPyyiltKS?usp=sharing)