Models: fine_tune_checkpoint

[fffupeng]

I want to try ssd_mobilenet_v1_pets with tensorflow object detection api. I have set every config files except

because i do not know how to set this:

When i download checkpoint from https://github.com/tensorflow/models/blob/master/object_detection/g3doc/detection_model_zoo.md, i just find graph.pbtxt,model.ckpt.data-00000-of-00001, model.ckpt.index, model.ckpt.meta and frozen_inference_graph.pb.There is no '.ckpt' files, so how can i set the following ???/model.ckpt file?
I didn't find any .ckpt file about ssd-mobilenet.
Is there any thing wrong with my download cheakpoint files?

[derekjchow]

Finetune checkpoint can point to a single model.ckpt file OR refer to a to a set of three files (.data-00000-of-00001, .index, .meta).

There's nothing wrong with your downloaded checkpoint files. Just use model.ckpt and replace "???" with the directory that contains the files.

[tatatodd]

Thanks for the explanation @derekjchow !

@fffupeng I'm assuming this resolves your problem. If not, just ping this thread and I'll re-open.

[fffupeng]

i have replace path with the dictionary ,just like this : fine_tune_checkpoint: "/home/fp/software/tensorflow/trainmodels/model.ckpt"
And i run
python /home/fp/software/tensorflow/models/object_detection/train.py
--pipeline_config_path=/home/fp/software/tensorflow/trainmodels/ssd_mobilenet_v1_pets.config
--train_dir=/home/fp/software/tensorflow
however it did not work ,make error as follows:
File "/home/fp/.local/lib/python2.7/site-packages/google/protobuf/text_format.py", line 652, in _MergeField
(message_descriptor.full_name, name))
google.protobuf.text_format.ParseError: 158:3 : Message type "object_detection.protos.Optimizer" has no field named "fine_tune_checkpoint".
Is there something wrong with my config? I just follow the guide in g3doc.

[scooter4j]

I'm confused about how to configure fine_tune_checkpoint for Google Cloud Machine Learning Engine. Using ssd_mobilenet_v2_coco, and like @fffupeng above I'm seeing the three model.ckpt.xxx files in the checkpoint download (and no model.ckpt file).

My pipeline config file for fine_tune_checkpoint is as follows:

.
.
.
      momentum_optimizer_value: 0.9
      decay: 0.9
      epsilon: 1.0
    }
  }
  fine_tune_checkpoint: "gs://training-model-smokey/model.ckpt"
  fine_tune_checkpoint_type:  "detection"

  # Note: The below line limits the training process to 200K steps, which we
  # empirically found to be sufficient enough to train the pets dataset. This
  # effectively bypasses the learning rate schedule (the learning rate will
.
.
.

but the Google CMLE fails looking for model.ckpt, as shown:

ERROR   2018-11-18 22:41:03 -0700   service     InvalidArgumentError: Unsuccessful TensorSliceReader constructor: Failed to get matching files on gs://training-model-smokey/model.ckpt: Not found: Error executing an HTTP request: HTTP response code 404 with body '{
ERROR   2018-11-18 22:41:03 -0700   service      "error": {
ERROR   2018-11-18 22:41:03 -0700   service       "errors": [
ERROR   2018-11-18 22:41:03 -0700   service        {
ERROR   2018-11-18 22:41:03 -0700   service         "domain": "global",
ERROR   2018-11-18 22:41:03 -0700   service         "reason": "notFound",
ERROR   2018-11-18 22:41:03 -0700   service         "message": "Not Found"
ERROR   2018-11-18 22:41:03 -0700   service        }
ERROR   2018-11-18 22:41:03 -0700   service       ],
ERROR   2018-11-18 22:41:03 -0700   service       "code": 404,
ERROR   2018-11-18 22:41:03 -0700   service       "message": "Not Found"
ERROR   2018-11-18 22:41:03 -0700   service      }
ERROR   2018-11-18 22:41:03 -0700   service     }
ERROR   2018-11-18 22:41:03 -0700   service     '
ERROR   2018-11-18 22:41:03 -0700   service          when reading gs://training-model-smokey/

How does one properly configure fine_tuning_checkpoint for Google CLME implementation?

[LiMengyang990726]

@fffupeng Hi, may I ask what is the link that you downloaded the existing checkpoint?