Skip to content
Adapted TensorFlow transfer learning code, specifying image lists rather than source directories
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

TF transfer learning on an image list

The adapted code provides the original TensorFlow (TF) hub example on transfer learning adjusted to ingest lists of image paths. Doing so, you avoid moving large amounts of image files into their respective working directories (per standard example).


Follow the standard installation instructions of all pre-requisites provided by the Tensorflow Hub transfer learning example.


Instead of using the or code, from the standard example, use the and code in this repository.

In addition you will need a CSV file specifying:

  • a files column with the full paths of your source images for training
  • a labels column which provides the label you associated with the content of the image
files, labels
/my/location/1.jpg, cat
/my/location/2.jpg, dog
..., ...

Other columns in this file will be ignored. Rows with NA as a label will be removed from the training / validation dataset.

retraining image labels

To retrain the model a new argument is introduced, image_file. The image_file argument specifies the location the csv file containing image and label information as outlined above. Other arugments remain unchanged.

# train the model
python ../python/ \
--image_file "/csv/file/location/image_files.csv" \
--flip_left_righ True \
--random_scale 10 \
--random_brightness 5 \
--how_many_training_steps 4000 \
--output_labels "~/your_labels.txt" \
--output_graph "~/your_graph.pb"

classifying images

Similar to the retraining code an image_list parameter was added to the original script. This list only requires a files column which specifies the location of images you want to see classified, all other columns are ignored.

# classify data
python ../python/ \
--image_list="../../data/crop_growth_labels.csv" \
--labels="~/your_labels.txt" \
--graph="~/your_graph.pb" \
You can’t perform that action at this time.