We are using the following packages for training CNNs using Caffe:
- DIGITS 6.1.1
- NVCaffe 0.17.1
- DIGITS image nvcr.io/nvidia/digits:19.06-caffe
However, selecting the “Compatible” option to save models in the old blob format does NOT add store_blobs_in_old_format in the solver.prototxt. Without this parameter, we cannot save models in a format compatible with BVLC caffe. This feature was released in DIGITS Release 18.05 (https://docs.nvidia.com/deeplearning/digits/digits-release-notes/rel_18.05.html) but is not working in 19.06. Any advice?