idealo/imageatm

Image classification for everyone.

imageatm's Issues

Add tests for components and utils

Upgrade to TensorFlow 2.0

Add docstrings for handlers

Make training component more transparent

Enforce strict type hinting and add it to Travis

The command mypy --config-file mypy.ini --strict imageatm should be included in Travis and only pass the build if all type hints are correct and complete.

Add tests for handlers

Further tests for both handlers:

• DataGenerator

• ImageClassifier

Add integration tests

setup.py fails, because dependencies

I get
error: PyYAML 5.2 is installed but PyYAML<=3.13,>=3.10 is required by {'awscli'}
and it seems a new version of awscli is to blame for that and indeed,
using

'awscli==1.11.18',

in setup.py works for me.

Evaluation report as jupyter notebook fails in cell 7

The following error appears when generating the evaluation report as jupyter notebook:

papermill.exceptions.PapermillExecutionError:
---------------------------------------------------------------------------
Exception encountered at "In [7]":
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-7-a62122680ff8> in <module>
----> 1 e._plot_classification_report(figsize=[max(5,e.n_classes*0.5), max(8,e.n_classes*0.8)])

AttributeError: 'Evaluation' object has no attribute '_plot_classification_report'

This is my config.yml:

evaluate:
  run: True
  report:
    create: True
    kernel_name: imageatm
    export_html: True
    export_pdf: False

Implement MixMatch strategy for semi-supervised learning

Replace os with pathlib in the entire project

For now not all elements of the project use pathlib.

Make image_dir and job_dir central configurations in config file.

• currently image_dir and job_dir can be set in config file for each component
• better to make them central configurations, i.e. put them on the same level as components in config file
• should be more user friendly, as we avoid complicated overwriting of image_dir and job_dir based on which components run first etc.

old format:

data_prep:
  run: True
  samples_file: sample_configfile.json
  image_dir: sample_dataset/
  job_dir: sample_jobdir/
  resize: True

train:
  run: True
  cloud: False

evaluate:
  run: True

cloud:
  cloud_tag: image_atm_example_tag
  provider: aws
  tf_dir: cloud/aws
  region: eu-west-1
  vpc_id: vpc-example
  instance_type: p2.xlarge
  bucket: s3://example-bucket
  destroy: True

new format:

image_dir: sample_dataset/
job_dir: sample_jobdir/

data_prep:
  run: True
  samples_file: sample_configfile.json
  resize: True

train:
  run: True
  cloud: False

evaluate:
  run: True

cloud:
  cloud_tag: image_atm_example_tag
  provider: aws
  tf_dir: cloud/aws
  region: eu-west-1
  vpc_id: vpc-example
  instance_type: p2.xlarge
  bucket: s3://example-bucket
  destroy: True

Add pdf and/or html output format for evaluation

Add terraform files to Python package.

Currently the terraform config files are not part of the python package and only available through cloning the repo.

Goal is to include these files in the package, such that no repo cloning is required to use the cloud resources.

https://stackoverflow.com/questions/56688232/including-non-python-files-in-a-python-package

Cloud fails when AWS VPC not set to default

If AWS VPC that is specified in config_file.yml under key vpc_id is not set to default VPC on AWS, the terraform apply process fails with the following error:

Error: Error applying plan:
1 error occurred:
	* module.ec2.aws_instance.image_atm: 1 error occurred:
	* aws_instance.image_atm: Error launching source instance: VPCIdNotSpecified: No default VPC for this user
	status code: 400, request id: xxx

Add Google Analytics to track documentation.

How can I make prediction on custom images by pre trained hdf5 model

Hi, First of all, I am new to this area, but I was able to create the hdf5 model as shown in the documentation. however, I want to test this model on different images using tensorflow. How can I do that?