satojkovic / DeepLogo

A brand logo detection system using tensorflow object detection API.
MIT License
512 stars 157 forks source link

Unable to process data from flickr_logos_27_dataset\flickr_logos_27_dataset_cropped_augmented_images\Lift\train.pickle #17

Open seevoid opened 6 years ago

seevoid commented 6 years ago

Hello !

I've tried to use this wonderful program to detect my own logos. For now i have just two logos wich represent a lift and an escalator. I change the files "flickr_logos_27_dataset_training_set_annotation.txt" and "flickr_logos_27_dataset_query_set_annotation.txt" to match with my samples. I also modified the python's file to set the number of classes to 2.

The crop_and_aug.py works well but when i launch gen_train_valid_test.py, after processes all the classes, I got the following error : "Unable to process data from flickr_logos_27_dataset\flickr_logos_27_dataset_cropped_augmented_images\Lift\train.pickle : could not broadcast input array from shape (576,32,64,3) into shape (2500,32,64,3) Traceback (most recent call last): File "gen_train_valid_test.py", line 192, in main() File "gen_train_valid_test.py", line 178, in main train_datasets, TRAIN_SIZE, VALID_SIZE) File "gen_train_valid_test.py", line 123, in merge_datasets valid_dataset[start_v:end_v, :, :, :] = valid_logo ValueError: could not broadcast input array from shape (576,32,64,3) into shape (2500,32,64,3)"

I would be greatful if you know how i can solve this issue :) Thanks !

[ EDIT 24/11/2017] After few hours looking for the origin of the issue, and i found that it come from my edit on the "flickr_logos_27_dataset_training_set_annotation.txt", and i don't really how to fill it properly. Do you know a simple way to easily make your own dataset ?

Regards Seevoid