lambdal / lambda-deep-learning-demo

Lambda Deep Learning Demos

Welcome to Lambda Lab's deep learning demo suite -- the place to find ready-to-use machine learnig models. We offer the following cool features:

• A curate of open-source, state-of-the-art models that cover major machine learning applications, including image classification, image segmentation, object detection, image style transfer, text classification and generation etc.

• Pure Tensorflow implementation. Efforts are made to keep the boilplate consistent across all demos.

• Examples of transfer learning and how to adapt the model to new data.

• Model serving

Check this documents for details.

Applications

---

Images Classification

Model	Dataset	Top 1 Accuracy	Pre-trained Model
ResNet32	CIFAR10	92%	Download
ResNet50 Fine-Tune	StanfordDogs	75.36%	Download
InceptionV4 Fine-Tune	StanfordDogs	92.4%	Download
NasNet-A-Large Fine-Tune	StanfordDogs	94.99%	Download

---

Images Segmentation

Model	Dataset	Accuracy	Pre-trained Model
FCN	CamVid	86.6%	Download
U-Net	CamVid	86.9%	Download

---

Object Detection

Model	Dataset	(AP) IoU=0.50:0.95	Pre-trained Model
SSD300	MSCOCO	21.9	Download
SSD500	MSCOCO	25.7	Download

---

Style Transfer

Model	Dataset	Pre-trained Model
Fast Neural Style	MSCOCO	Download

---

Text Generation

Model	Dataset	Pre-trained Model
Char RNN	Shakespeare	Download
Word RNN	Shakespeare	Download
Word RNN + Glove	Shakespeare	Download

---

Text Classification

Model	Dataset	Classification Accuracy	Pre-trained Model
LSTM	IMDB	85.2%	Download
LSTM + Glove	IMDB	86.1%	Download
Transfer Learning + BERT	IMDB	92.2%	Download

---

Citation

If you use our code in your research or wish to refer to the examples, please cite with:

@misc{lambdalabs2018demo,
  title={Lambda Labs Deep Learning Demos},
  author={Lambda Labs, inc.},
  howpublished={\url{https://github.com/lambdal/lambda-deep-learning-demo}},
  year={2018}
}

lambda-deep-learning-demo's Issues

TypeError: Cannot create initializer for non-floating point type.

i am getting this error on tf1.12 gpu

AttributeError: 'module' object has no attribute 'random'

I am following the Object Detection Tutorial, trying to Train SSD from scratch on MSCOCO

python demo/object_detection.py \
--mode=train --model_dir=~/demo/model/ssd300_mscoco \
--network=ssd300 --augmenter=ssd_augmenter --batch_size_per_gpu=16 --epochs=100 \
--dataset_dir=/mnt/data/data/mscoco --num_classes=81 --resolution=300 \
--feature_net=vgg_16_reduced --feature_net_path=demo/model/VGG_16_reduce/VGG_16_reduce.p \
train_args --learning_rate=0.001 --optimizer=momentum --piecewise_boundaries=60,80 \
--piecewise_lr_decay=1.0,0.1,0.01 --dataset_meta=train2014,valminusminival2014 \
--callbacks=train_basic,train_loss,train_speed,train_summary \
--skip_l2_loss_vars=l2_norm_scaler --summary_names=loss,learning_rate,class_losses,bboxes_losses

Which version of Tensorflow do I need to use for this tutorial? I tried both 1.0.0 & 1.11.0 in Python2.7 virtual environments.

>>> tensorflow.version
'1.11.0'

loading annotations into memory...
Done (t=9.26s)
creating index...
index created!
loading annotations into memory...
Done (t=7.51s)
creating index...
index created!
Traceback (most recent call last):
  File "demo/object_detection.py", line 134, in <module>
    main()
  File "demo/object_detection.py", line 130, in main
    runner.run()
  File "./source/runner/runner.py", line 146, in run
    self.create_graph()
  File "./source/runner/parameter_server_runner.py", line 132, in create_graph
    reduced_ops = self.replicate_graph()
  File "./source/runner/parameter_server_runner.py", line 73, in replicate_graph
    batch = self.inputter.input_fn()
  File "./source/inputter/object_detection_mscoco_inputter.py", line 231, in input_fn
    num_parallel_calls=12)
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/data/ops/dataset_ops.py", line 1007, in map
    return ParallelMapDataset(self, map_func, num_parallel_calls)
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/data/ops/dataset_ops.py", line 2248, in __init__
    super(ParallelMapDataset, self).__init__(input_dataset, map_func)
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/data/ops/dataset_ops.py", line 2216, in __init__
    map_func, "Dataset.map()", input_dataset)
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/data/ops/dataset_ops.py", line 1473, in __init__
    self._function.add_to_graph(ops.get_default_graph())
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/framework/function.py", line 479, in add_to_graph
    self._create_definition_if_needed()
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/framework/function.py", line 335, in _create_definition_if_needed
    self._create_definition_if_needed_impl()
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/framework/function.py", line 344, in _create_definition_if_needed_impl
    self._capture_by_value, self._caller_device)
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/framework/function.py", line 865, in func_graph_from_py_func
    outputs = func(*func_graph.inputs)
  File "/usr/lib/python2.7/dist-packages/tensorflow/python/data/ops/dataset_ops.py", line 1411, in tf_data_structured_function_wrapper
    ret = func(*nested_args)
  File "./source/inputter/object_detection_mscoco_inputter.py", line 230, in <lambda>
    image_id, file_name, classes, boxes),
  File "./source/inputter/object_detection_mscoco_inputter.py", line 208, in parse_fn
    speed_mode=False)
  File "./source/augmenter/ssd_augmenter.py", line 532, in augment
    speed_mode=speed_mode)
  File "./source/augmenter/ssd_augmenter.py", line 418, in preprocess_for_train
    image, boxes = random_zoom_out(image, boxes)
  File "./source/augmenter/ssd_augmenter.py", line 283, in random_zoom_out
    uniform_random = tf.random.uniform([], 0, 1.0, seed=seed)
AttributeError: 'module' object has no attribute 'random'

How to perform inference on a video

I am trying to run a video and perform inference. I am unable to do that.

Attempted to Train SSD from scratch on MSCOCO

Following the tutorial here: https://lambda-deep-learning-demo.readthedocs.io/en/latest/tutorial/ssd.html and I tried to train SSD and I get an error in the ssd_common.py script:

Traceback (most recent call last):
  File "demo/image/object_detection.py", line 134, in <module>
    main()
  File "demo/image/object_detection.py", line 108, in main
    "source.network." + modeler_config.network)
  File "/home/srdc/anaconda3/lib/python3.7/importlib/__init__.py", line 127, in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
  File "<frozen importlib._bootstrap>", line 1006, in _gcd_import
  File "<frozen importlib._bootstrap>", line 983, in _find_and_load
  File "<frozen importlib._bootstrap>", line 967, in _find_and_load_unlocked
  File "<frozen importlib._bootstrap>", line 677, in _load_unlocked
  File "<frozen importlib._bootstrap_external>", line 728, in exec_module
  File "<frozen importlib._bootstrap>", line 219, in _call_with_frames_removed
  File "./source/network/ssd300.py", line 26, in <module>
    INPUT_DIM)
  File "./source/network/detection/ssd_common.py", line 102, in get_anchors
    for ratio in xrange(min_size_ratio, max_size_ratio + 1, step):
NameError: name 'xrange' is not defined

What should I do to get the training step complete?