Merge ../object-locator

Former-commit-id: e2eb49fb47bb142a89815aa9710c6edbc24c913c

.gitignore

@@ -0,0 +1,107 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+.hypothesis/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+.static_storage/
+.media/
+local_settings.py
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# pyenv
+.python-version
+
+# celery beat schedule file
+celerybeat-schedule
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+
+# vim
+*.swp

CHANGELOG.md

@@ -0,0 +1,144 @@
+### Changelog
+
+#### 1.5.0
+ * [b81e40f](../../commit/b81e40f) - __(Javier Ribera)__ + explanation in argparse
+ * [298ad6c](../../commit/298ad6c) - __(Javier Ribera)__ added tif as valid extension for imgs
+ * [218545e](../../commit/218545e) - __(Javier Ribera)__ random seed for initial split of validation dataset
+ * [6110944](../../commit/6110944) - __(Javier Ribera)__ automatically detect XML or CSV, and move plant stuff to separate module
+ * [5d0fd32](../../commit/5d0fd32) - __(Javier Ribera)__ new intermediate output w/ crosses on original img
+ * [2c03135](../../commit/2c03135) - __(Javier Ribera)__ organize: move some dirs into intermediate dir
+ * [54f9c7a](../../commit/54f9c7a) - __(Javier Ribera)__ fix: ignore nan thresholds when plotting kde
+ * [b0a1cf3](../../commit/b0a1cf3) - __(Javier Ribera)__ split subrow_grid_location_yx into y and x
+ * [2f2c42d](../../commit/2f2c42d) - __(Javier Ribera)__ fix: saving empty CSV
+ * [3b8b0f5](../../commit/3b8b0f5) - __(Javier Ribera)__ fix: forgot to create output dir when not evaluating
+ * [d7eaf62](../../commit/d7eaf62) - __(Javier Ribera)__ fix: crash when there is not GT (just inference)
+ * [c321949](../../commit/c321949) - __(Javier Ribera)__ use XML instead of CSV for inference
+ * [abbbeb1](../../commit/abbbeb1) - __(Javier Ribera)__ fix: build dictionary with ints instead of tensors
+ * [8853d76](../../commit/8853d76) - __(Javier Ribera)__ update to pytorch 1.0.0
+ * [04a582a](../../commit/04a582a) - __(Javier Ribera)__ fix: convert str to float when chekcing argparse
+ * [c87cd4b](../../commit/c87cd4b) - __(Javier Ribera)__ force some arguments to be strictly positive
+ * [de05268](../../commit/de05268) - __(Javier Ribera)__ argparse: dont show default if empty str
+ * [c2b27c7](../../commit/c2b27c7) - __(Javier Ribera)__ mention which args are mandatory, and which optional
+ * [4ac1f58](../../commit/4ac1f58) - __(Javier Ribera)__ update visdom to newer version
+ * [4243c12](../../commit/4243c12) - __(Javier Ribera)__ move to XML, allow partial GT in XML, and 80-20 autosplit w/ XML
+ * [c6d19ff](../../commit/c6d19ff) - __(Javier Ribera)__ fix visdom not connecting w/ workaround
+ * [c85cbfe](../../commit/c85cbfe) - __(Javier Ribera)__ make visdom optional, port modifiable, and print visdom hostname
+ * [7692159](../../commit/7692159) - __(Javier Ribera)__ fix (estimated map must be 1D)
+ * [50cf1bc](../../commit/50cf1bc) - __(Javier Ribera)__ mention # images for training/validation
+ * [2fee7e6](../../commit/2fee7e6) - __(Javier Ribera)__ go back to UNet
+ * [502daa8](../../commit/502daa8) - __(Javier Ribera)__ full EM
+ * [5aabb5a](../../commit/5aabb5a) - __(Javier Ribera)__ new option to replace optimizer
+
+#### 1.4.0
+ * [86cc171](../../commit/86cc171) - __(Javier Ribera)__ removed non-used models
+ * [b8c6ba1](../../commit/b8c6ba1) - __(Javier Ribera)__ read from XML v0.4
+ * [101cfa3](../../commit/101cfa3) - __(Javier Ribera)__ fix: get_image_size in results script
+ * [70cd619](../../commit/70cd619) - __(Javier Ribera)__ max_ahd=diagonal in metrics_from_results script
+ * [2de70e3](../../commit/2de70e3) - __(Javier Ribera)__ new script to make metrics CSV from results CSV
+ * [c5b487c](../../commit/c5b487c) - __(Javier Ribera)__ trying CSRNet
+ * [a52f835](../../commit/a52f835) - __(Javier Ribera)__ merge BMM density and threshold in same plot
+ * [fd3afd8](../../commit/fd3afd8) - __(Javier Ribera)__ cap BMM plots at density=8 instead of 50
+ * [40f0094](../../commit/40f0094) - __(Javier Ribera)__ heatmap on top of original image during inference
+ * [e5c7498](../../commit/e5c7498) - __(Javier Ribera)__ force R=1 when object count is known
+ * [21d4462](../../commit/21d4462) - __(Javier Ribera)__ use default types of fixed # of objects
+ * [bed824d](../../commit/bed824d) - __(Javier Ribera)__ nicer plots
+ * [bf9d0cb](../../commit/bf9d0cb) - __(Javier Ribera)__ allow [] in taus in arguments
+ * [a1c9209](../../commit/a1c9209) - __(Javier Ribera)__ move legend up in the plots
+ * [5562a13](../../commit/5562a13) - __(Javier Ribera)__ go back use SmoothL1Loss for regression
+ * [e67753f](../../commit/e67753f) - __(Javier Ribera)__ new script to draw circles on top of images of a dataset
+ * [7ce6002](../../commit/7ce6002) - __(Javier Ribera)__ remove unused variables
+ * [91006d0](../../commit/91006d0) - __(Javier Ribera)__ fix: p=-1 by default to prevent NaNs
+ * [4f5b7a4](../../commit/4f5b7a4) - __(Javier Ribera)__ fix: prevent NaNs by not using activ in last layers of U-Net
+ * [4645147](../../commit/4645147) - __(Javier Ribera)__ fix: use seed in random transforms too
+ * [9890290](../../commit/9890290) - __(Javier Ribera)__ paint red haircrosses on estimated pts during validation
+ * [9747046](../../commit/9747046) - __(Javier Ribera)__ go back to use Otsu thresholding
+ * [386e4f3](../../commit/386e4f3) - __(Javier Ribera)__ use seed when shuffling dataset at the beginning
+ * [d79a1d7](../../commit/d79a1d7) - __(Javier Ribera)__ also paint white circles on top of training labels
+ * [6504a7c](../../commit/6504a7c) - __(Javier Ribera)__ paint red circles on top of estimated points and send to Visdom
+ * [61e70d4](../../commit/61e70d4) - __(Javier Ribera)__ send heatmaps on top of original to visdom
+ * [6a57704](../../commit/6a57704) - __(Javier Ribera)__ shuffle dataset at the beginning
+ * [166af89](../../commit/166af89) - __(Javier Ribera)__ fix: use mean of second Beta in BMM thresholding
+ * [034b96f](../../commit/034b96f) - __(Javier Ribera)__ use BMM instead of Otsu thresholding during validation
+ * [6812ff3](../../commit/6812ff3) - __(Javier Ribera)__ fix: tau is a float, not an int
+ * [c2fdeed](../../commit/c2fdeed) - __(Javier Ribera)__ fix: skip validation every val_freq times
+ * [362ffca](../../commit/362ffca) - __(Javier Ribera)__ new script to find optimal learning rate (order of magnitude)
+ * [16b5d94](../../commit/16b5d94) - __(Javier Ribera)__ new option --force to overwrite output testing results
+ * [4464750](../../commit/4464750) - __(Javier Ribera)__ track and log running average loss
+ * [d988c42](../../commit/d988c42) - __(Javier Ribera)__ also send avg loss to visdom
+ * [335cc16](../../commit/335cc16) - __(Javier Ribera)__ fix: logger not accepting numbers
+ * [0f54e01](../../commit/0f54e01) - __(Javier Ribera)__ fix: no need for model.train() for every batch (now faster)
+ * [a68072d](../../commit/a68072d) - __(Javier Ribera)__ added copyright info of BMM fitting
+ * [c1aab76](../../commit/c1aab76) - __(Javier Ribera)__ nicer plots
+ * [d2e9490](../../commit/d2e9490) - __(Javier Ribera)__ fix: multiple radii as input
+ * [a393e06](../../commit/a393e06) - __(Javier Ribera)__ use Beta Mixture Model-based thresholding with tau=-2
+ * [a6db2b8](../../commit/a6db2b8) - __(Javier Ribera)__ fit spherical GMM so it runs faster
+ * [3a37888](../../commit/3a37888) - __(Javier Ribera)__ subsample mask points randomly so GMM fitting is faster
+ * [4c65e51](../../commit/4c65e51) - __(Javier Ribera)__ fix: validation was not being done
+ * [ccc414b](../../commit/ccc414b) - __(Javier Ribera)__ label otsu thresholding in metric plots
+ * [01b01f5](../../commit/01b01f5) - __(Javier Ribera)__ fix: take only some radii if too many to plot
+ * [0147511](../../commit/0147511) - __(Javier Ribera)__ rename variable windows->window_ids
+ * [1276bee](../../commit/1276bee) - __(Javier Ribera)__ fix: visdom not show validation because reusing window_ids
+ * [becbfd0](../../commit/becbfd0) - __(Javier Ribera)__ clearer print msg
+ * [413f526](../../commit/413f526) - __(Javier Ribera)__ fix: show always same # of decimals of tau
+ * [b7091d1](../../commit/b7091d1) - __(Javier Ribera)__ fix: scale estimated map before saving as img
+ * [0c2bde3](../../commit/0c2bde3) - __(Javier Ribera)__ fix: multiple taus in argparse
+ * [75d0bdc](../../commit/75d0bdc) - __(Javier Ribera)__ use AMSgrad, the "convergence fix" for Adam
+ * [22dd9b5](../../commit/22dd9b5) - __(Javier Ribera)__ always printing losses should not be there besides for debugging
+ * [02af554](../../commit/02af554) - __(Javier Ribera)__ p=9 by default
+ * [7247aa3](../../commit/7247aa3) - __(Javier Ribera)__ new 2nd term in the cost function
+ * [b0db956](../../commit/b0db956) - __(Javier Ribera)__ clarify README instructions
+ * [ed72736](../../commit/ed72736) - __(Javier Ribera)__ no need to save checkpoint at the end of each epoch if no validation
+ * [38f762c](../../commit/38f762c) - __(Javier Ribera)__ smaller scatter markers
+ * [ea968e7](../../commit/ea968e7) - __(Javier Ribera)__ 50 taus instead of 100
+ * [288288b](../../commit/288288b) - __(Javier Ribera)__ added copyright notices to all files
+ * [9e1b82b](../../commit/9e1b82b) - __(Javier Ribera)__ script is now part of the package
+ * [0ae815c](../../commit/0ae815c) - __(Javier Ribera)__ fix doc: setup.py said the package was inference-only
+ * [8840a16](../../commit/8840a16) - __(Javier Ribera)__ able to send visualiz to remote Visdom server
+ * [f5b1490](../../commit/f5b1490) - __(Javier Ribera)__ conda environment: missing ballpark dependency
+ * [e1d8af6](../../commit/e1d8af6) - __(Javier Ribera)__ wrap clustering into a function
+ * [018157c](../../commit/018157c) - __(Javier Ribera)__ fix: undo accidental removal of line in commit d78ecf6
+ * [ecfd007](../../commit/ecfd007) - __(Javier Ribera)__ max AHD=max dist instead of inf during training
+ * [d78ecf6](../../commit/d78ecf6) - __(Javier Ribera)__ encapsulate thresholding into a function
+ * [a726245](../../commit/a726245) - __(Javier Ribera)__ corner case: avoid crash when tau = -1 only
+ * [570dc74](../../commit/570dc74) - __(Javier Ribera)__ better var names, torch0.4 optimizations, and use centroids_wrt_orig
+ * [61ad18b](../../commit/61ad18b) - __(Javier Ribera)__ mark Otsu result separate in the metric plots
+ * [dedb81b](../../commit/dedb81b) - __(Javier Ribera)__ fix: not reporting MAE
+ * [83b3628](../../commit/83b3628) - __(Javier Ribera)__ use item() from pythorch0.4
+ * [f7d53c0](../../commit/f7d53c0) - __(Javier Ribera)__ forgot Normalizer class
+ * [43fc661](../../commit/43fc661) - __(Javier Ribera)__ use Otsu thresholding in testing
+ * [65bde7c](../../commit/65bde7c) - __(Javier Ribera)__ show # params during testing
+ * [23b4c41](../../commit/23b4c41) - __(Javier Ribera)__ normalize centroids in a new class
+ * [63f743f](../../commit/63f743f) - __(Javier Ribera)__ show number of parameters
+ * [58a377c](../../commit/58a377c) - __(Javier Ribera)__ fix corner case when img has no object at all
+ * [77a2f88](../../commit/77a2f88) - __(Javier Ribera)__ use non-default port number in visdom
+ * [5bab050](../../commit/5bab050) - __(Javier Ribera)__ one painted output per threshold value
+ * [082c07a](../../commit/082c07a) - __(Javier Ribera)__ tau=1 => otsu thresholding
+ * [507642c](../../commit/507642c) - __(Javier Ribera)__ change default visdom environment name
+ * [ec0d9b2](../../commit/ec0d9b2) - __(Javier Ribera)__ also show r and R^2 on validation in visdom
+ * [ad7d18e](../../commit/ad7d18e) - __(Javier Ribera)__ fix: dimension mismatch
+ * [8f9824b](../../commit/8f9824b) - __(Javier Ribera)__ avoid matplotlib warning
+ * [aab2727](../../commit/aab2727) - __(Javier Ribera)__ corner case in WHD when there are no GT pts
+ * [6349981](../../commit/6349981) - __(Javier Ribera)__ use default types
+ * [311225d](../../commit/311225d) - __(Javier Ribera)__ seed is 0 now
+ * [15030d7](../../commit/15030d7) - __(Javier Ribera)__ also scale image during testing
+ * [c1d82c9](../../commit/c1d82c9) - __(Javier Ribera)__ fix matplotlib warning by closing unused figures
+ * [113cc3f](../../commit/113cc3f) - __(Javier Ribera)__ also compute r and R2 metrics
+ * [15b03a7](../../commit/15b03a7) - __(Javier Ribera)__ plotting metrics inside py package, and plot when locating too
+ * [4de5506](../../commit/4de5506) - __(Javier Ribera)__ new script to plot multiple metrics from the results of a CSV
+ * [7be5a71](../../commit/7be5a71) - __(Javier Ribera)__ use otsu thresholding
+ * [4218727](../../commit/4218727) - __(Javier Ribera)__ fix CSV dataset loading
+ * [0f3a3dc](../../commit/0f3a3dc) - __(Javier Ribera)__ put additional info in output CSV
+
+#### 1.3.1
+ * [9969e3b](../../commit/9969e3b) - __(Javier Ribera)__ v1.3.1
+ * [6937c18](../../commit/6937c18) - __(Javier Ribera)__ no more wheels
+ * [988a5b1](../../commit/988a5b1) - __(Javier Ribera)__ allow empty checkpoint directory
+ * [191fb3c](../../commit/191fb3c) - __(Javier Ribera)__ show error when dataset directory is empty
+ * [c2c05c5](../../commit/c2c05c5) - __(Javier Ribera)__ object-location -> object-locator
+
+#### 1.3.0
+ * [b53d485](../../commit/b53d485) - __(Javier Ribera)__ fix environment
+ * [1156a67](../../commit/1156a67) - __(Javier Ribera)__ environment.yml for Windows
+ * [c0d5490](../../commit/c0d5490) - __(Javier Ribera)__ change deprecated class to remove warning
+ * [859be9d](../../commit/859be9d) - __(Javier Ribera)__ better argparse help
+

COPYRIGHT.txt

@@ -0,0 +1,9 @@
+Copyright &copyright 2018 The Board of Trustees of Purdue University.
+All rights reserved.
+
+This source code is not to be distributed or modified 
+without the written permission of Edward J. Delp at Purdue University.
+Contact information: ace@ecn.purdue.edu
+
+Last Modified: 09/21/2018
+

Dockerfile

@@ -0,0 +1,28 @@
+FROM nvidia/cuda:8.0-cudnn6-devel-ubuntu16.04
+
+# Install conda
+RUN apt-get -qq update && \
+    apt-get -qq -y install curl bzip2 qtbase5-dev libgtk2.0-0\
+    && curl -sSL https://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -o /tmp/miniconda.sh \
+    && bash /tmp/miniconda.sh -bfp /usr/local \
+    && rm -rf /tmp/miniconda.sh \
+    && conda update conda \
+    && apt-get -qq -y remove curl bzip2 \
+    && apt-get -qq -y autoremove \
+    && apt-get autoclean \
+    && rm -rf /var/lib/apt/lists/* /var/log/dpkg.log \
+    && conda clean --all --yes
+
+COPY . /object_locator
+WORKDIR /object_locator
+
+# Create conda environment with all the dependencies
+RUN [ "conda", "env", "create", "--file", "environment.yml" ]
+
+# Install object-locator python package inside the conda environment
+RUN [ "/bin/bash", "-c", "source activate object-locator && python setup.py install" ]
+
+# Prepare entrypoint, which just calls the object-locator
+RUN echo '#!/bin/bash \n source activate object \n python -m object-locator "$@"' > /object_locator/entrypoint.bash
+RUN chmod u+x "/object_locator/entrypoint.bash"
+ENTRYPOINT ["/object_locator/entrypoint.bash"]

README.md

@@ -1,7 +1,8 @@
 # Locating Objects Without Bounding Boxes
 PyTorch code for https://arxiv.org/pdf/1806.07564.pdf
 
-Code will be uploaded by mid April, 2019.
+![Object localization](https://i.postimg.cc/bNcFr9Pf/collage3x3.png)
+
 
 ## Citing this work
 ```
@@ -15,3 +16,62 @@ Code will be uploaded by mid April, 2019.
 }
 ```
 
+## Datasets
+  The datasets used in the paper can be downloaded from:
+  - [Mall dataset](http://personal.ie.cuhk.edu.hk/~ccloy/downloads_mall_dataset.html)
+  - [Pupil dataset](http://www.ti.uni-tuebingen.de/Pupil-detection.1827.0.html)
+  - [Plant dataset](https://engineering.purdue.edu/~sorghum/dataset-plant-centers-2016)
+
+## Installation
+Use conda to recreate the environment provided with the code:
+
+<pre>
+conda env create -f environment.yml
+</pre>
+
+and install the tool:
+
+<pre>
+pip install .
+</pre>
+
+## Usage  
+Activate the environment:
+<pre>
+conda activate object-locator
+</pre>
+
+Run this to get help (usage instructions):
+<pre>
+python -m object-locator.locate -h
+python -m object-locator.train -h
+</pre>
+
+Example:
+<pre>
+python -m object-locator.locate \
+       --dataset DIRECTORY \
+       --out DIRECTORY \
+       --model CHECKPOINTS \
+       --evaluate \
+       --no-gpu \
+       --radius 5
+
+python -m object-locator.train \
+       --train-dir TRAINING_DIRECTORY \
+       --batch-size 32 \
+       --env-name sorghum \
+       --lr 1e-3 \
+       --val-dir TRAINING_DIRECTORY \
+       --optim Adam \
+       --save saved_model.ckpt
+</pre>
+
+## Uninstall
+
+<pre>
+conda deactivate object-locator
+conda env remove --name object-locator
+</pre>
+
+

environment.yml

@@ -0,0 +1,31 @@
+name: object-locator
+channels:
+    - pytorch
+    - conda-forge
+    - defaults
+dependencies:
+    - imageio=2.3.0
+    - ipdb=0.11
+    - ipython=6.3.1
+    - ipython_genutils=0.2.0
+    - matplotlib=2.2.2
+    - numpy=1.14.3
+    - opencv=3.4.1
+    - pandas=0.22.0
+    - parse=1.8.2
+    - pip=9.0.3
+    - python=3.6.5
+    - python-dateutil=2.7.2
+    - scikit-image=0.13.1
+    - scikit-learn=0.19.1
+    - scipy=1.0.1
+    - setuptools=39.1.0
+    - tqdm=4.23.1
+    - xmltodict=0.11.0
+    - pytorch=1.0.0
+    - pip:
+        - ballpark==1.4.0
+        - visdom==0.1.8.5
+        - peterpy
+        - torchvision==0.2.1
+