no tf_efficientnet_b3_ns_s8 model in geffnet
cv-lab-x opened this issue · comments
hi, thanks for your great work!
there's no tf_efficientnet_b3_ns_s8 model in geffnet, only tf_efficientnet_b3_ns,
looking forward to your reply, thanks @spyflying
Sorry for our previous negligence. We have made several modifications to the original efficientnet for task adaptation. By adjusting the downsampling strides, we obtain tf_efficientnet_b3_ns_s8 based on the original tf_efficientnet_b3_ns. We have included the modified geffnet module. Please re-clone the updated repo.
Sorry for our previous negligence. We have made several modifications to the original efficientnet for task adaptation. By adjusting the downsampling strides, we obtain tf_efficientnet_b3_ns_s8 based on the original tf_efficientnet_b3_ns. We have included the modified geffnet module. Please re-clone the updated repo.
it works, thanks for your quick reply!
Another question: how did you create the data_lists/traning.txt and data_lists/validation.txt, cause there're not data_lists folder in openlane data, and i found that can not find some files in openlane datasets in validation.txt eg: validation/segment-271338158136329280_2541_070_2561_070_with_camera_labels/150869126825189000
The data lists are generated based on the processed pickle files, i.e., the training.txt includes files with 1 or more lanes in the training set, and the validation.txt includes all the files in the validation set. You can freely generate the training.txt by collecting all the file names in your cache_dense/training folder.
For the missing validation file, I guess it may be due to the different versions of Openlane. This file is not included in the annotations of Openlane v1.2 compared with the original Openlane annotations. You can remove this line from validation.txt and I think there should be little difference in the validation results.