Retraining pruned model and conv size error

I am trying to retrain the pruned model.
My training spec before pruning is default_spec_resnet34.log (3.6 KB) .

My pruning command is
tlt-prune -m /workspace/tlt-experiments/FasterRCNN/resnet34/numplate_frcnn_resnet34.epoch15.tlt -o /workspace/tlt-experiments/FasterRCNN/resnet34/prune/resnet34_prune.tlt -k ******** -eq union -pth 0.5

The pruned model is retrain with the spec file retrain_spec_resnet34.log (3.6 KB) .

I have error as
File “/usr/local/bin/tlt-train-g1”, line 8, in
File “./common/”, line 33, in main
File “./faster_rcnn/scripts/”, line 56, in main
File “./faster_rcnn/models/”, line 244, in build_or_resume_model
File “./faster_rcnn/models/”, line 339, in load_weights
File “/usr/local/lib/python2.7/dist-packages/keras/engine/”, line 1163, in load_weights
File “./faster_rcnn/patched_keras/”, line 113, in load_weights_from_hdf5_group_by_name
ValueError: Layer #1 (named “conv1”), weight <tf.Variable ‘conv1/kernel:0’ shape=(7, 7, 3, 64) dtype=float32_ref> has shape (7, 7, 3, 64), but the saved weight has shape (16, 3, 7, 7).

What could be wrong?

Now is ok. I changed to another spec file ‘default_spec_resnet18_retrain_spec.txt’