Object detection training from_detection_checkpoint doesn't keep step number

System information

• What is the top-level directory of the model you are using: models/research/object_detection
• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): no
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): os x
• TensorFlow installed from (source or binary): source
• TensorFlow version (use command below): v1.14.0-0-g87989f6959 1.14.0
• Bazel version (if compiling from source): NA
• CUDA/cuDNN version: NA
• GPU model and memory: NA
• Exact command to reproduce:

python /content/models/research/object_detection/model_main.py \
    --pipeline_config_path=/content/resnet101.config \
    --model_dir=/content/output \
    --alsologtostderr

Describe the problem

When fine-tuning from a detection checkpoint, even with from_detection_checkpoint: true, the training process restarts at step 0.

Example, with:

  fine_tune_checkpoint: "restnet101/model.ckpt-1234"
  from_detection_checkpoint: true

Training will load the checkpoint without issue but will start at step 0, not at step 1234. I am pretty sure this behaviour changed "recently", do you know what might be causing this? Did I get something wrong in my config?