tensorflow / tensorflow/models

Fine Tunning Model at step 0 detects nothing

Open
#8,686 0 comments 0 reactions 3 assignees View on GitHub

@pkulzc is already working on this.

Since Jun 17, 2020.

models:research:odapi type:support
Dominant language
Python
Stars
77.7k
Forks
44.8k
PR merge metrics
No merged PRs in 30d

Description

I have been tring to fine tunning a pre-trained model with my own dataset.

I have changed "fine_tune_checkpoint to point" to the downloaded ckpt and "from_detection_checkpoint" to true

However, it seems that the training process wiped out all the pre-trained weights and started from scratch, because when I use the checkpont generated at step 0, it cannot detect anything.

The pretrained model I have selected is faster_rcnn_inception_v2_coco.
I have also tried rfcn_resnet101_coco, same problem.

Another thing I have noticed is that the downloaded ckpt is about 52099 KB, but the checkpoints during training is about twice the size (100596KB) and after exported to inference_graph, the size reduced to again to similar size (50337KB)...
Can anyone please shed a light on this size change as well?
Thank you in advance!

System Info:
Ubuntu 18.04.2 LTS
Python 3.6.9
Tensorflow 1.12.0
CUDA 9.0
CPU i9-9920X
GPU TITAN RTX

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.