aktwelve / cocosynth Goto Github PK

Hi,

I'am currently trying to detect weeds in sugar beet field. I used akTwelve cocosynth codes to generate my dataset (10,000 training images and 1,000 validation images). Thanks a lot, those codes are amazing.
Now I'm trying to train the mask-rcnn model adapted from Matterport repo and an issue occurred when I give the path of the coco_instances.json files ("Create the Training and Validation Datasets", in cell 11).

As you can see it seems to be an error in "load_data" while reading "image_id" in the json file.
In those file I noticed that at the end of the file lots of commas are "messy" (for big datasets like the training one). I generated small dataset (with "clean" jsons files) and the issue in the code was the same, so it seems not to be the issue.

I precise that the training code work perfectly with the jsons files from the "box_dataset_synthetic_complete" that akTwelve made.

I found no help anywhere on the web and my teacher do not understand the error to.
Does anyone have an idea to fix this trouble and help me?

Thanks a lot for helping me!

Best
Maxime

The link is broken and I think I am missing a step cloning the repo

hey hi, i had a problem with COCO Instances JSON (coco_json_utils.py) while i'm creating a json file i got these error AttributeError: 'MultiPolygon' object has no attribute 'exterior' .
can you please help me with this?

?

Good night, first congratulations on the work. I was following the tutorial that taught how to train MaskR CNN with my own dataset, I did all the previous steps, generated the images with "image_composition.py", changed the number of classes and dimensions of the images in the model.

However, in the Create the Training and Validation Datasets section, the code is giving the following error:

"KeyError Traceback (most recent call last)
in
1 dataset_train = CocoLikeDataset ()
2 dataset_train.load_data ('../ datasets / box_dataset_synthetic / train / coco_instances.json',
----> 3 '../datasets/box_dataset_synthetic/train/images')
4 dataset_train.prepare ()
5

in load_data (self, annotation_json, images_dir)
49
50 image_path = os.path.abspath (os.path.join (images_dir, image_file_name))
---> 51 image_annotations = annotations [image_id]
52
53 # Add the image using the base method from utils.Dataset

KeyError: 654 "

detail: the name of the dataset is correct, I changed it to the same name as yours.

When I put your coco_instances.json it works, but in mine it doesn't work, I compared both and I couldn't detect differences.

Can you please help me?

Thank you very much

How to make coco dataset from scratch complete tutorial

https://www.youtube.com/watch?v=MrXlCD3UfZY

Hi, congratulations on your work, today I tried to create images using 'image_composition.py', I create the same directory tree that you recommend, but when I start image_composition.py the errors occur:

python3 ./python/image_composition.py --input_dir ./datasets/challenge_dataset/input --output_dir ./datasets/challenge_dataset/output --count 10 --width 480 --height 360
Traceback (most recent call last):
File "./python/image_composition.py", line 484, in
image_comp.main(args)
File "./python/image_composition.py", line 458, in main
self._validate_and_process_args(args)
File "./python/image_composition.py", line 148, in _validate_and_process_args
self._validate_and_process_input_directory()
File "./python/image_composition.py", line 179, in _validate_and_process_input_directory
assert self.foregrounds_dir is not None, 'foregrounds sub-directory was not found in the input_dir'
AttributeError: 'ImageComposition' object has no attribute 'foregrounds_dir'

Thank you!!

Hi Adam,
I followed your course on udemy, it was great.
Unfortunately, when I tried running the mask rcnn training, it kept hanging.
Please is this an issue because I am using tensorflow v2?
in the course I have seen that you mentioned cloning matterport_mask_rcnn.
However I have seen you have another repo mask_rcnn on your git which of the two should I use please.
Your feedback would be appreciated.

On the page https://github.com/akTwelve/cocosynth

hie,
actually I'm new to object detection. can you please help me out from these errors

1. while viewing coco dataset
2. while loading train dataset
   
   

Thank you.

Is there a way to extend the Coco_Image_Viewer with a function to display keypoints?

i have five png data and put them in foregrounds folder. but i got 'no valid foregrounds were found' error. give me a hand

I am curious where you obtained the weed images? From an open dataset?

For my use case, I do not want any foreground images to overlap when being placed in a background image. Is there a way to prevent this from happening?

Thanks for the code!
In your Udemy course, do you use Keras(TF) or Pytorch? I would like to use Detectron2(Pytorch). Does it work as well or it doesn't really matter which framework I use? Thanks!