mini-pytorch-examples

How to Train Model

Download data:

$ cd pytorch-cls
$ wget https://download.pytorch.org/tutorial/hymenoptera_data.zip

$ cd pytorch-seg/data
$ wget http://www.doc.ic.ac.uk/~ahanda/nyu_test_rgb.tgz
$ wget http://www.doc.ic.ac.uk/~ahanda/nyu_train_rgb.tgz
$ wget https://github.com/ankurhanda/nyuv2-meta-data/raw/master/test_labels_13/nyuv2_test_class13.tgz
$ wget https://github.com/ankurhanda/nyuv2-meta-data/raw/master/train_labels_13/nyuv2_train_class13.tgz

Run the training script:

$ python3 train.py

Here's an example of how to pass in various arguments are provided in the script pytorch-seg/train-model-exp1.sh:

# pytorch-seg/train-model-exp1.sh

python train.py \
--dataroot "/media/shrek/work/datasets/shuran-mini-pytorch-examples/" \
--file_list "./data/datalist" \
--batchSize 4 \
--shuffle True \
--phase train \
--num_epochs 10 \
--imsize 224 \
--logs_path "logs/exp3"

The checkpoints will be stored in the pytorch-seg/checkpoints folder.

How to Visualize the data via Tensorboard

Support has been added for visualizing data through Tensorboard via the pytorch package TensorBoardX.

Install TensorBoardX:

$ pip install git+https://github.com/lanpa/tensorboardX

While the model is training, run:

$ tensorboard --logdir pytorch-seg/logs

Viewing model graph

You can also view a graph of the model. Note that bilinear upsampling is not yet supported:

Changelog

v1.1

• Added tensorboard visualization
• Cleaned up some deprecated bilinear upsampling functions
• Added documentation

v1.0

• Added code

Dependency

Check readme within pytorch-normals/