Aspect_ratios and scales in retinanet_spec

  aspect_ratios_global: "[1.0, 2.0, 0.5]"
  scales: "[0.045, 0.09, 0.2, 0.4, 0.55, 0.7]"

Working fine with above values , when i try to increase scale value > 6 or aspect_ratios_global >3 , Got error…

Also got error when i change ‘aspect_ratios_global’ into ‘aspect_ratios’

See https://docs.nvidia.com/metropolis/TLT/tlt-getting-started-guide/index.html#retinanet_config_retinanet for more details.

aspect_ratios_global: “[1.0, 2.0, 0.5]”
scales: “[0.05,0.1, 0.25, 0.423, 0.55, 0.7,0.9]”

Getting following error when the length of scales >6 . But i need to add scale length >6

2020-08-29 05:12:01,391 [INFO] iva.retinanet.scripts.train: Loading pretrained weights. This may take a while…
Traceback (most recent call last):
File “/usr/local/bin/tlt-train-g1”, line 8, in
sys.exit(main())
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/715c8bafe7816f3bb6f309cd506049bb/execroot/ai_infra/bazel-out/k8-py3-fastbuild/bin/magnet/packages/iva/buil
d_wheel.runfiles/ai_infra/iva/common/magnet_train.py”, line 48, in main
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/715c8bafe7816f3bb6f309cd506049bb/execroot/ai_infra/bazel-out/k8-py3-fastbuild/bin/magnet/packages/iva/buil
d_wheel.runfiles/ai_infra/iva/retinanet/scripts/train.py”, line 248, in main
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/715c8bafe7816f3bb6f309cd506049bb/execroot/ai_infra/bazel-out/k8-py3-fastbuild/bin/magnet/packages/iva/buil
d_wheel.runfiles/ai_infra/iva/retinanet/scripts/train.py”, line 106, in run_experiment
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/715c8bafe7816f3bb6f309cd506049bb/execroot/ai_infra/bazel-out/k8-py3-fastbuild/bin/magnet/packages/iva/buil
d_wheel.runfiles/ai_infra/iva/retinanet/utils/model_io.py”, line 71, in load_model_as_pretrain
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/715c8bafe7816f3bb6f309cd506049bb/execroot/ai_infra/bazel-out/k8-py3-fastbuild/bin/magnet/packages/iva/buil
d_wheel.runfiles/ai_infra/iva/retinanet/builders/model_builder.py”, line 67, in build
File “/home/vpraveen/.cache/dazel/_dazel_vpraveen/715c8bafe7816f3bb6f309cd506049bb/execroot/ai_infra/bazel-out/k8-py3-fastbuild/bin/magnet/packages/iva/buil
d_wheel.runfiles/ai_infra/iva/retinanet/architecture/retinanet.py”, line 259, in retinanet
ValueError: It must be either scales is None or len(scales) == 6, but len(scales) == 7.

In retinanet, the n_predictor conv layers in the network is 5.
The length of scales must be either None or n_predictor_layers+1.