• Network Type : Object detection & Efficientdet
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here) Used docker nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf2.11.0
May I know what is Efficientdet_tf2’s version. Is it D5 or D6?
According to this picture from one of Nvidia’s webinart, Efficientdet-D7 has highest accuracy.
In tf2 docker nvcr.io/nvidia/tao/tao-toolkit:5.0.0-tf2.11.0,
The Efficientdet network supports efficientdet-d0 to efficientdet-d5 according to
https://github.com/NVIDIA/tao_tensorflow2_backend/blob/main/nvidia_tao_tf2/cv/efficientdet/model/model_builder.py#L93-L105.
The classification network can support efficientnet-b0 to efficientnet-b7 according to
https://github.com/NVIDIA/tao_tensorflow2_backend/blob/26e599687749f7726fb3f796ccafe10520315e2d/nvidia_tao_tf2/cv/classification/model/model_builder.py.
Where can I change to train with efficientdet-d5?
Is it in spec_train.yaml?
Is pretrained model same for all versions?
Pre-trained is downloaded from here.
There is only one version of pretrained model.
Please refer to Open Images Pre-trained EfficientDet - NVIDIA Docs → TAO Pretrained EfficientDet | NVIDIA NGC.
You can set it in “checkpoint” in the train yaml file.
Wow a lot. Why I couldn’t find them before :(
I have this error.
I am using efficientdet-tf2.
The error is
OSError: SavedModel file does not exist at: /workspace/Nyan/tao_source_codes_v5.0.0/notebooks/tao_launcher_starter_kit/efficientdet_tf2/pretrained/{saved_model.pbtxt|saved_model.pb}
Now testing with tf1 docker
Yes, tf1 docker and efficientdet-tf1 are ok.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.
