feat: Add dataset script

Signed-off-by: Diwank Singh Tomer <[email protected]>

dataset_infos.json

@@ -1,6 +1,8 @@
-{"lld": {
+{
+  "lld": {
     "description": "Designing a logo for a new brand is a lengthy and tedious back-and-forth process between a designer and a client. In this paper we explore to what extent machine learning can solve the creative task of the designer. For this, we build a dataset -- LLD -- of 600k+ logos crawled from the world wide web. Training Generative Adversarial Networks (GANs) for logo synthesis on such multi-modal data is not straightforward and results in mode collapse for some state-of-the-art methods. We propose the use of synthetic labels obtained through clustering to disentangle and stabilize GAN training. We are able to generate a high diversity of plausible logos and we demonstrate latent space exploration techniques to ease the logo design task in an interactive manner. Moreover, we validate the proposed clustered GAN training on CIFAR 10, achieving state-of-the-art Inception scores when using synthetic labels obtained via clustering the features of an ImageNet classifier. GANs can cope with multi-modal data by means of synthetic labels achieved through clustering, and our results show the creative potential of such techniques for logo synthesis and manipulation.",
-    "citation": "@misc{sage2017logodataset, author={Sage, Alexander and Agustsson, Eirikur and Timofte, Radu and Van Gool, Luc}, title = {LLD - Large Logo Dataset - version 0.1}, year = {2017}, howpublished = "\url{https://data.vision.ee.ethz.ch/cvl/lld}"}",
+    "citation": "@misc{sage2017logodataset, author={Sage, Alexander and Agustsson, Eirikur and Timofte, Radu and Van Gool, Luc}, title = {LLD - Large Logo Dataset - version 0.1}, year = {2017}, howpublished = url{https://data.vision.ee.ethz.ch/cvl/lld}}",
     "homepage": "https://data.vision.ee.ethz.ch/sagea/lld/",
     "config_name": "labeled"
-}}
+  }
+}

download.sh

@@ -0,0 +1,3 @@
+#!/usr/bin/env bash
+
+wget -O raw/LLD-logo.hdf5 https://data.vision.ee.ethz.ch/sagea/lld/data/LLD-logo.hdf5

lld/loader.py

@@ -7,13 +7,15 @@ import numpy as np
 import PIL.Image as Image
 
 script_dir = os.path.dirname(__file__)
-datafile_path = os.path.join(script_dir, "../data/LLD-logo.hdf5")
+datafile_path = os.path.join(script_dir, "../raw/LLD-logo.hdf5")
 
 with h5py.File(datafile_path, "r") as throwaway:
     samples_count: int = len(throwaway["data"])
 
 
-def gen_samples(labels: list[str] = ["data", "meta_data/names"]):
+def gen_samples(
+    labels: list[str] = ["data", "meta_data/names"], datafile_path: str = datafile_path
+):
 
     # open hdf5 file
     with h5py.File(datafile_path, "r") as hdf5_file:

lld/preprocess.py

@@ -9,18 +9,20 @@ import numpy as np
 import pandas as pd
 from tqdm.asyncio import trange
 
-from .loader import gen_samples, samples_count
+from .loader import datafile_path, gen_samples, samples_count
 from .crawler import run
 
 script_dir = os.path.dirname(__file__)
 outfile_path = os.path.join(script_dir, "../data/lld-processed.h5")
 
 
-async def gen_processor(batch_size: int, limit: int):
+async def gen_processor(
+    batch_size: int, limit: int, datafile_path: str = datafile_path
+):
     count = min(limit, samples_count)
     batch_size = min(limit, batch_size)
 
-    samples = gen_samples()
+    samples = gen_samples(datafile_path=datafile_path)
     steps = count // batch_size
 
     for step in trange(steps):
@@ -42,24 +44,33 @@ async def gen_processor(batch_size: int, limit: int):
             yield data
 
 
-async def preprocess(batch_size: int = 100, limit: int = samples_count + 1):
+async def preprocess(
+    batch_size: int = 100,
+    limit: int = samples_count + 1,
+    datafile_path: str = datafile_path,
+):
 
     columns = ["images", "description", "name"]
 
-    processor = gen_processor(batch_size, limit)
+    processor = gen_processor(batch_size, limit, datafile_path=datafile_path)
 
     chunk_size = 1000
     async with stream.chunks(processor, chunk_size).stream() as chunks:
         async for chunk in chunks:
             df_chunk = pd.DataFrame(chunk, columns=columns)
-            df_chunk.to_hdf(
-                outfile_path, "data", data_columns=columns, mode="a"
-            )
+            df_chunk.to_hdf(outfile_path, "data", data_columns=columns, mode="a")
 
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser()
 
+    parser.add_argument(
+        "--datafile_path",
+        help="Path to downloaded archive",
+        type=str,
+        default=datafile_path,
+    )
+
     parser.add_argument(
         "--limit",
         help="Limit to total records processed",

lld_logos.py

@@ -0,0 +1,86 @@
+"""Dataset class for LLD dataset."""
+
+import datasets as ds
+import pandas as pd
+from sklearn.model_selection import train_test_split
+
+
+_HOMEPAGE = "https://huggingface.co/datasets/diwank/lld"
+_LICENSE = "MIT"
+
+_DESCRIPTION = """
+Designing a logo for a new brand is a lengthy and tedious back-and-forth process between a designer and a client. In this paper we explore to what extent machine learning can solve the creative task of the designer. For this, we build a dataset -- LLD -- of 600k+ logos crawled from the world wide web. Training Generative Adversarial Networks (GANs) for logo synthesis on such multi-modal data is not straightforward and results in mode collapse for some state-of-the-art methods. We propose the use of synthetic labels obtained through clustering to disentangle and stabilize GAN training. We are able to generate a high diversity of plausible logos and we demonstrate latent space exploration techniques to ease the logo design task in an interactive manner. Moreover, we validate the proposed clustered GAN training on CIFAR 10, achieving state-of-the-art Inception scores when using synthetic labels obtained via clustering the features of an ImageNet classifier. GANs can cope with multi-modal data by means of synthetic labels achieved through clustering, and our results show the creative potential of such techniques for logo synthesis and manipulation.
+"""
+
+_CITATION = """
+@misc{sage2017logodataset,
+author={Sage, Alexander and Agustsson, Eirikur and Timofte, Radu and Van Gool, Luc},
+title = {LLD - Large Logo Dataset - version 0.1},
+year = {2017},
+"""
+
+_URL = "https://huggingface.co/datasets/diwank/lld/resolve/main/data/lld-processed.h5"
+
+
+class LLD(ds.GeneratorBasedBuilder):
+    """LLD Images dataset."""
+
+    def _info(self):
+        print("_info(self):")
+        import pdb; pdb.set_trace()
+        return ds.DatasetInfo(
+            description=_DESCRIPTION,
+            features=ds.Features(
+                {
+                    "image": ds.Sequence(feature=ds.Image()),
+                    "description": ds.Value("string"),
+                }
+            ),
+            supervised_keys=("image", "description"),
+            homepage=_HOMEPAGE,
+            citation=_CITATION,
+            license=_LICENSE,
+        )
+
+    def _split_generators(self, dl_manager):
+        print("_split_generators(self, dl_manager):")
+        import pdb; pdb.set_trace()
+        # Load dataframe
+        use_local = os.environ.get("USE_LOCAL")
+        archive_path = (
+            "./data/lld-processed.h5" if use_local else dl_manager.download(_URL)
+        )
+        df = pd.read_hdf(archive_path)
+
+        X = df.pop("description")
+        y = df.pop("images")
+        X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2)
+
+        return [
+            ds.SplitGenerator(
+                name=ds.Split.TRAIN,
+                gen_kwargs={
+                    "description": X_train,
+                    "images": y_train,
+                },
+            ),
+            ds.SplitGenerator(
+                name=ds.Split.TEST,
+                gen_kwargs={
+                    "description": X_test,
+                    "images": y_test,
+                },
+            ),
+        ]
+
+    def _generate_examples(self, description, images):
+        print("_generate_examples(self, description, images):")
+        import pdb; pdb.set_trace()
+        """Generate images and description splits."""
+
+        for i, (desc, imgs) in enumerate(zip(description.values, images.values)):
+            for img in imgs:
+                yield i, {
+                    "image": {"bytes": img},
+                    "description": desc,
+                }

poetry.lock

@@ -264,6 +264,7 @@ multiprocess = "*"
 numpy = ">=1.17"
 packaging = "*"
 pandas = "*"
+Pillow = {version = ">=6.2.1", optional = true, markers = "extra == \"vision\""}
 pyarrow = ">=6.0.0"
 requests = ">=2.19.0"
 responses = "<0.19"
@@ -535,6 +536,14 @@ MarkupSafe = ">=2.0"
 [package.extras]
 i18n = ["Babel (>=2.7)"]
 
+[[package]]
+name = "joblib"
+version = "1.1.0"
+description = "Lightweight pipelining with Python functions"
+category = "main"
+optional = false
+python-versions = ">=3.6"
+
 [[package]]
 name = "locket"
 version = "1.0.0"
@@ -900,6 +909,37 @@ urllib3 = ">=1.25.10"
 [package.extras]
 tests = ["pytest (>=4.6)", "coverage (>=6.0.0)", "pytest-cov", "pytest-localserver", "flake8", "types-mock", "types-requests", "mypy"]
 
+[[package]]
+name = "scikit-learn"
+version = "1.1.1"
+description = "A set of python modules for machine learning and data mining"
+category = "main"
+optional = false
+python-versions = ">=3.8"
+
+[package.dependencies]
+joblib = ">=1.0.0"
+numpy = ">=1.17.3"
+scipy = ">=1.3.2"
+threadpoolctl = ">=2.0.0"
+
+[package.extras]
+benchmark = ["matplotlib (>=3.1.2)", "pandas (>=1.0.5)", "memory-profiler (>=0.57.0)"]
+docs = ["matplotlib (>=3.1.2)", "scikit-image (>=0.14.5)", "pandas (>=1.0.5)", "seaborn (>=0.9.0)", "memory-profiler (>=0.57.0)", "sphinx (>=4.0.1)", "sphinx-gallery (>=0.7.0)", "numpydoc (>=1.2.0)", "Pillow (>=7.1.2)", "sphinx-prompt (>=1.3.0)", "sphinxext-opengraph (>=0.4.2)"]
+examples = ["matplotlib (>=3.1.2)", "scikit-image (>=0.14.5)", "pandas (>=1.0.5)", "seaborn (>=0.9.0)"]
+tests = ["matplotlib (>=3.1.2)", "scikit-image (>=0.14.5)", "pandas (>=1.0.5)", "pytest (>=5.0.1)", "pytest-cov (>=2.9.0)", "flake8 (>=3.8.2)", "black (>=22.3.0)", "mypy (>=0.770)", "pyamg (>=4.0.0)", "numpydoc (>=1.2.0)"]
+
+[[package]]
+name = "scipy"
+version = "1.8.1"
+description = "SciPy: Scientific Library for Python"
+category = "main"
+optional = false
+python-versions = ">=3.8,<3.11"
+
+[package.dependencies]
+numpy = ">=1.17.3,<1.25.0"
+
 [[package]]
 name = "six"
 version = "1.16.0"
@@ -964,6 +1004,14 @@ category = "main"
 optional = false
 python-versions = ">=2.7, !=3.0.*, !=3.1.*, !=3.2.*, !=3.3.*, !=3.4.*"
 
+[[package]]
+name = "threadpoolctl"
+version = "3.1.0"
+description = "threadpoolctl"
+category = "main"
+optional = false
+python-versions = ">=3.6"
+
 [[package]]
 name = "tomli"
 version = "2.0.1"
@@ -1086,8 +1134,8 @@ heapdict = "*"
 
 [metadata]
 lock-version = "1.1"
-python-versions = "^3.10"
-content-hash = "a2e4f349bb1aaa25793b817b43a1c0bb6c008857aa2cc30df73baf5e048dfa22"
+python-versions = ">=3.10,<3.11"
+content-hash = "ebfc659f0dac7f7b70f1533416f7e76ea3e1b56ce81bc072fe196357c22d82c5"
 
 [metadata.files]
 aiodns = []
@@ -1326,6 +1374,7 @@ jedi = [
     {file = "jedi-0.18.1.tar.gz", hash = "sha256:74137626a64a99c8eb6ae5832d99b3bdd7d29a3850fe2aa80a4126b2a7d949ab"},
 ]
 jinja2 = []
+joblib = []
 locket = []
 markupsafe = []
 matplotlib-inline = [
@@ -1570,6 +1619,8 @@ responses = [
     {file = "responses-0.18.0-py3-none-any.whl", hash = "sha256:15c63ad16de13ee8e7182d99c9334f64fd81f1ee79f90748d527c28f7ca9dd51"},
     {file = "responses-0.18.0.tar.gz", hash = "sha256:380cad4c1c1dc942e5e8a8eaae0b4d4edf708f4f010db8b7bcfafad1fcd254ff"},
 ]
+scikit-learn = []
+scipy = []
 six = [
     {file = "six-1.16.0-py2.py3-none-any.whl", hash = "sha256:8abb2f1d86890a2dfb989f9a77cfcfd3e47c2a354b01111771326f8aa26e0254"},
     {file = "six-1.16.0.tar.gz", hash = "sha256:1e61c37477a1626458e36f7b1d82aa5c9b094fa4802892072e49de9c60c4c926"},
@@ -1585,6 +1636,7 @@ stack-data = [
 ]
 tables = []
 tblib = []
+threadpoolctl = []
 tomli = []
 toolz = []
 tornado = []

pyproject.toml

@@ -6,10 +6,10 @@ authors = ["Diwank Singh Tomer <[email protected]>"]
 license = "MIT"
 
 [tool.poetry.dependencies]
-python = "^3.10"
+python = ">=3.10,<3.11"
 numpy = "^1.23.1"
 h5py = "^3.7.0"
-datasets = "^2.3.2"
+datasets = {extras = ["vision"], version = "^2.3.2"}
 Pillow = "^9.2.0"
 requests = "^2.28.1"
 bs4 = "^0.0.1"
@@ -22,6 +22,8 @@ tables = "^3.7.0"
 pyarrow = "^8.0.0"
 modin = {extras = ["dask"], version = "^0.15.2"}
 aiostream = "^0.4.4"
+scipy = "^1.8.1"
+scikit-learn = "^1.1.1"
 
 [tool.poetry.dev-dependencies]
 ipython = "^8.4.0"