Merge branch 'main' into dl-and-pp-as-parquet

Author: lhoestq

datasets/crd3/README.md

@@ -55,9 +55,6 @@ paperswithcode_id: crd3
 - **Repository:** [CRD3 repository](https://github.com/RevanthRameshkumar/CRD3)
 - **Paper:** [Storytelling with Dialogue: A Critical Role Dungeons and Dragons Dataset](https://www.aclweb.org/anthology/2020.acl-main.459/)
 - **Point of Contact:** [More Information Needed](https://github.com/huggingface/datasets/blob/master/CONTRIBUTING.md#how-to-contribute-to-the-dataset-cards)
-- **Size of downloaded dataset files:** 279.93 MB
-- **Size of the generated dataset:** 4020.33 MB
-- **Total amount of disk used:** 4300.25 MB
 
 ### Dataset Summary
 
@@ -69,6 +66,7 @@ collaboration and spoken interaction. For each dialogue, there are a large numbe
 and semantic ties to the previous dialogues.
 
 ### Supported Tasks and Leaderboards
+
 `summarization`: The dataset can be used to train a model for abstractive summarization. A [fast abstractive summarization-RL](https://github.com/ChenRocks/fast_abs_rl) model was presented as a baseline, which achieves ROUGE-L-F1 of 25.18.
 
 ### Languages
@@ -79,13 +77,8 @@ The text in the dataset is in English, as spoken by actors on The Critical Role
 
 ### Data Instances
 
-#### default
-
-- **Size of downloaded dataset files:** 279.93 MB
-- **Size of the generated dataset:** 4020.33 MB
-- **Total amount of disk used:** 4300.25 MB
-
 An example of 'train' looks as follows.
+
 ```
 {
     "alignment_score": 3.679936647415161,
@@ -105,7 +98,6 @@ An example of 'train' looks as follows.
 
 The data fields are the same among all splits.
 
-#### default
 - `chunk`: a `string` feature.
 - `chunk_id`: a `int32` feature.
 - `turn_start`: a `int32` feature.
@@ -120,7 +112,7 @@ The data fields are the same among all splits.
 
 | name  | train |validation| test  |
 |-------|------:|---------:|------:|
-|default|26,232|   3,470|4,541|
+|default|38,969|   6,327|7,500|
 
 ## Dataset Creation
 
@@ -180,19 +172,16 @@ This work is licensed under a [Creative Commons Attribution-ShareAlike 4.0 Inter
 
 ### Citation Information
 
-```
-
+```bibtex
 @inproceedings{
 title = {Storytelling with Dialogue: A Critical Role Dungeons and Dragons Dataset},
 author = {Rameshkumar, Revanth  and Bailey, Peter},
 year = {2020},
 publisher = {Association for Computational Linguistics},
 conference = {ACL}
 }
-
 ```
 
-
 ### Contributions
 
 Thanks to [@thomwolf](https://github.com/thomwolf), [@lhoestq](https://github.com/lhoestq), [@mariamabarham](https://github.com/mariamabarham), [@lewtun](https://github.com/lewtun) for adding this dataset.

datasets/crd3/crd3.py

@@ -45,11 +45,11 @@
 and semantic ties to the previous dialogues.
 """
 
-_URL = "https://github.com/RevanthRameshkumar/CRD3/archive/master.zip"
+_URL = "https://huggingface.co/datasets/crd3/resolve/72bffe55b4d5bf19b530d3e417447b3384ba3673/data/aligned%20data.zip"
 
 
 def get_train_test_dev_files(files, test_split, train_split, dev_split):
-    test_files = dev_files = train_files = []
+    test_files, dev_files, train_files = [], [], []
     for file in files:
         filename = os.path.split(file)[1].split("_")[0]
         if filename in test_split:
@@ -88,20 +88,22 @@ def _info(self):
         )
 
     def _split_generators(self, dl_manager):
-        path = dl_manager.download_and_extract(_URL)
-        test_file = os.path.join(path, "CRD3-master", "data", "aligned data", "test_files")
-        train_file = os.path.join(path, "CRD3-master", "data", "aligned data", "train_files")
-        dev_file = os.path.join(path, "CRD3-master", "data", "aligned data", "val_files")
+        root = dl_manager.download_and_extract(_URL)
+        path = os.path.join(root, "aligned data")
+
+        test_file = os.path.join(path, "test_files")
+        train_file = os.path.join(path, "train_files")
+        dev_file = os.path.join(path, "val_files")
         with open(test_file, encoding="utf-8") as f:
             test_splits = [file.replace("\n", "") for file in f.readlines()]
 
         with open(train_file, encoding="utf-8") as f:
             train_splits = [file.replace("\n", "") for file in f.readlines()]
         with open(dev_file, encoding="utf-8") as f:
             dev_splits = [file.replace("\n", "") for file in f.readlines()]
-        c2 = "CRD3-master/data/aligned data/c=2"
-        c3 = "CRD3-master/data/aligned data/c=3"
-        c4 = "CRD3-master/data/aligned data/c=4"
+        c2 = "c=2"
+        c3 = "c=3"
+        c4 = "c=4"
         files = [os.path.join(path, c2, file) for file in sorted(os.listdir(os.path.join(path, c2)))]
         files.extend([os.path.join(path, c3, file) for file in sorted(os.listdir(os.path.join(path, c3)))])
         files.extend([os.path.join(path, c4, file) for file in sorted(os.listdir(os.path.join(path, c4)))])

datasets/crd3/dataset_infos.json

@@ -1 +1 @@
-{"default": {"description": "\nStorytelling with Dialogue: A Critical Role Dungeons and Dragons Dataset.\nCritical Role is an unscripted, live-streamed show where a fixed group of people play Dungeons and Dragons, an open-ended role-playing game.\nThe dataset is collected from 159 Critical Role episodes transcribed to text dialogues, consisting of 398,682 turns. It also includes corresponding\nabstractive summaries collected from the Fandom wiki. The dataset is linguistically unique in that the narratives are generated entirely through player\ncollaboration and spoken interaction. For each dialogue, there are a large number of turns, multiple abstractive summaries with varying levels of detail,\nand semantic ties to the previous dialogues.\n", "citation": "\n@inproceedings{\ntitle = {Storytelling with Dialogue: A Critical Role Dungeons and Dragons Dataset},\nauthor = {Rameshkumar, Revanth  and Bailey, Peter},\nyear = {2020},\npublisher = {Association for Computational Linguistics},\nconference = {ACL}\n}\n ", "homepage": "https://github.com/RevanthRameshkumar/CRD3", "license": "", "features": {"chunk": {"dtype": "string", "id": null, "_type": "Value"}, "chunk_id": {"dtype": "int32", "id": null, "_type": "Value"}, "turn_start": {"dtype": "int32", "id": null, "_type": "Value"}, "turn_end": {"dtype": "int32", "id": null, "_type": "Value"}, "alignment_score": {"dtype": "float32", "id": null, "_type": "Value"}, "turns": {"feature": {"names": {"dtype": "string", "id": null, "_type": "Value"}, "utterances": {"dtype": "string", "id": null, "_type": "Value"}, "number": {"dtype": "int32", "id": null, "_type": "Value"}}, "length": -1, "id": null, "_type": "Sequence"}}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "crd3", "config_name": "default", "version": {"version_str": "0.0.0", "description": null, "major": 0, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 318560673, "num_examples": 52796, "dataset_name": "crd3"}, "test": {"name": "test", "num_bytes": 318560673, "num_examples": 52796, "dataset_name": "crd3"}, "validation": {"name": "validation", "num_bytes": 318560673, "num_examples": 52796, "dataset_name": "crd3"}}, "download_checksums": {"https://github.com/RevanthRameshkumar/CRD3/archive/master.zip": {"num_bytes": 294222220, "checksum": "c77a937394f265735ba54b32a7a051f77a97d264c74b0535dee77ef9791815b5"}}, "download_size": 294222220, "post_processing_size": null, "dataset_size": 955682019, "size_in_bytes": 1249904239}}
+{"default": {"description": "\nStorytelling with Dialogue: A Critical Role Dungeons and Dragons Dataset.\nCritical Role is an unscripted, live-streamed show where a fixed group of people play Dungeons and Dragons, an open-ended role-playing game.\nThe dataset is collected from 159 Critical Role episodes transcribed to text dialogues, consisting of 398,682 turns. It also includes corresponding\nabstractive summaries collected from the Fandom wiki. The dataset is linguistically unique in that the narratives are generated entirely through player\ncollaboration and spoken interaction. For each dialogue, there are a large number of turns, multiple abstractive summaries with varying levels of detail,\nand semantic ties to the previous dialogues.\n", "citation": "\n@inproceedings{\ntitle = {Storytelling with Dialogue: A Critical Role Dungeons and Dragons Dataset},\nauthor = {Rameshkumar, Revanth  and Bailey, Peter},\nyear = {2020},\npublisher = {Association for Computational Linguistics},\nconference = {ACL}\n}\n ", "homepage": "https://github.com/RevanthRameshkumar/CRD3", "license": "", "features": {"chunk": {"dtype": "string", "id": null, "_type": "Value"}, "chunk_id": {"dtype": "int32", "id": null, "_type": "Value"}, "turn_start": {"dtype": "int32", "id": null, "_type": "Value"}, "turn_end": {"dtype": "int32", "id": null, "_type": "Value"}, "alignment_score": {"dtype": "float32", "id": null, "_type": "Value"}, "turns": [{"names": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "utterances": {"feature": {"dtype": "string", "id": null, "_type": "Value"}, "length": -1, "id": null, "_type": "Sequence"}, "number": {"dtype": "int32", "id": null, "_type": "Value"}}]}, "post_processed": null, "supervised_keys": null, "task_templates": null, "builder_name": "crd3", "config_name": "default", "version": {"version_str": "0.0.0", "description": null, "major": 0, "minor": 0, "patch": 0}, "splits": {"train": {"name": "train", "num_bytes": 236605152, "num_examples": 38969, "dataset_name": "crd3"}, "test": {"name": "test", "num_bytes": 40269203, "num_examples": 7500, "dataset_name": "crd3"}, "validation": {"name": "validation", "num_bytes": 41543528, "num_examples": 6327, "dataset_name": "crd3"}}, "download_checksums": {"https://huggingface.co/datasets/crd3/resolve/72bffe55b4d5bf19b530d3e417447b3384ba3673/data/aligned%20data.zip": {"num_bytes": 117519820, "checksum": "c66bd9f7848bcd514a35c154edd2fc874f1a3076876d8bd7208bf3caf4b7fb0b"}}, "download_size": 117519820, "post_processing_size": null, "dataset_size": 318417883, "size_in_bytes": 435937703}}

datasets/crd3/dummy/0.0.0/dummy_data.zip

@@ -39,7 +39,11 @@ Both parameter values default to 1000, which can be expensive if you are storing
 
 ## Data augmentation
 
-🤗 Datasets can apply data augmentations from any library or package to your dataset. This guide will use the transforms from [torchvision](https://pytorch.org/vision/stable/transforms.html).
+🤗 Datasets can apply data augmentations from any library or package to your dataset.
+
+### Image Classification
+
+First let's see how you can transform image classification datasets. This guide will use the transforms from [torchvision](https://pytorch.org/vision/stable/transforms.html).
 
 <Tip>
 
@@ -88,3 +92,163 @@ Now you can take a look at the augmented image by indexing into the `pixel_value
     <img class="block dark:hidden" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/image_process_jitter.png">
     <img class="hidden dark:block" src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/image_process_jitter.png"/>
 </div>
+
+### Object Detection
+
+Object detection models identify something in an image, and object detection datasets are used for applications such as autonomous driving and detecting natural hazards like wildfire. This guide will show you how to apply transformations to an object detection dataset following the [tutorial](https://albumentations.ai/docs/examples/example_bboxes/) from [Albumentations](https://albumentations.ai/docs/).
+
+To run these examples, make sure you have up-to-date versions of `albumentations` and `cv2` installed:
+
+```
+pip install -U albumentations opencv-python
+```
+
+In this example, you'll use the [`cppe-5`](https://huggingface.co/datasets/cppe-5) dataset for identifying medical personal protective equipment (PPE) in the context of the COVID-19 pandemic.
+
+Load the dataset and take a look at an example:
+
+```py
+from datasets import load_dataset
+
+>>> ds = load_dataset("cppe-5")
+>>> example = ds['train'][0]
+>>> example
+{'height': 663,
+ 'image': <PIL.JpegImagePlugin.JpegImageFile image mode=RGB size=943x663 at 0x7FC3DC756250>,
+ 'image_id': 15,
+ 'objects': {'area': [3796, 1596, 152768, 81002],
+  'bbox': [[302.0, 109.0, 73.0, 52.0],
+   [810.0, 100.0, 57.0, 28.0],
+   [160.0, 31.0, 248.0, 616.0],
+   [741.0, 68.0, 202.0, 401.0]],
+  'category': [4, 4, 0, 0],
+  'id': [114, 115, 116, 117]},
+ 'width': 943}
+```
+
+The dataset has the following fields:
+
+- `image`: PIL.Image.Image object containing the image.
+- `image_id`: The image ID.
+- `height`: The image height.
+- `width`: The image width.
+- `objects`: A dictionary containing bounding box metadata for the objects in the image:
+  - `id`: The annotation id.
+  - `area`: The area of the bounding box.
+  - `bbox`: The object's bounding box (in the [coco](https://albumentations.ai/docs/getting_started/bounding_boxes_augmentation/#coco) format).
+  - `category`: The object's category, with possible values including `Coverall (0)`, `Face_Shield (1)`, `Gloves (2)`, `Goggles (3)` and `Mask (4)`.
+
+You can visualize the `bboxes` on the image using some internal torch utilities. To do that, you will need to reference the [`~datasets.ClassLabel`] feature associated with the category IDs so you can look up the string labels:
+
+    
+```py
+>>> import torch
+>>> from torchvision.ops import box_convert
+>>> from torchvision.utils import draw_bounding_boxes
+>>> from torchvision.transforms.functional import pil_to_tensor, to_pil_image
+
+>>> categories = ds['train'].features['objects'].feature['category']
+
+>>> boxes_xywh = torch.tensor(example['objects']['bbox'])
+>>> boxes_xyxy = box_convert(boxes_xywh, 'xywh', 'xyxy')
+>>> labels = [categories.int2str(x) for x in example['objects']['category']]
+>>> to_pil_image(
+...     draw_bounding_boxes(
+...         pil_to_tensor(example['image']),
+...         boxes_xyxy,
+...         colors="red",
+...         labels=labels,
+...     )
+... )
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/nateraw/documentation-images/resolve/main/visualize_detection_example.png">
+</div>
+
+
+With `albumentations`, you can apply transforms that will affect the image while also updating the `bboxes` accordingly. In this case, the image is resized to (480, 480), flipped horizontally, and brightened. 
+
+`albumentations` expects the image to be in BGR format, not RGB, so you'll have to convert the image before applying the transform.
+
+```py
+>>> import albumentations as A
+>>> import numpy as np
+
+>>> transform = A.Compose([
+...     A.Resize(480, 480),
+...     A.HorizontalFlip(p=1.0),
+...     A.RandomBrightnessContrast(p=1.0),
+... ], bbox_params=A.BboxParams(format='coco',  label_fields=['category']))
+
+>>> # RGB PIL Image -> BGR Numpy array
+>>> image = np.flip(np.array(example['image']), -1)
+>>> out = transform(
+...     image=image,
+...     bboxes=example['objects']['bbox'],
+...     category=example['objects']['category'],
+... )
+```
+
+Now when you visualize the result, the image should be flipped, but the `bboxes` should still be in the right places.
+
+```py
+>>> image = torch.tensor(out['image']).flip(-1).permute(2, 0, 1)
+>>> boxes_xywh = torch.stack([torch.tensor(x) for x in out['bboxes']])
+>>> boxes_xyxy = box_convert(boxes_xywh, 'xywh', 'xyxy')
+>>> labels = [categories.int2str(x) for x in out['category']]
+>>> to_pil_image(
+...     draw_bounding_boxes(
+...         image,
+...         boxes_xyxy,
+...         colors='red',
+...         labels=labels
+...     )
+... )
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/nateraw/documentation-images/resolve/main/visualize_detection_example_transformed.png">
+</div>
+
+Create a function to apply the transform to a batch of examples:
+
+```py
+>>> def transforms(examples):
+...     images, bboxes, categories = [], [], []
+...     for image, objects in zip(examples['image'], examples['objects']):
+...         image = np.array(image.convert("RGB"))[:, :, ::-1]
+...         out = transform(
+...             image=image,
+...             bboxes=objects['bbox'],
+...             category=objects['category']
+...         )
+...         images.append(torch.tensor(out['image']).flip(-1).permute(2, 0, 1))
+...         bboxes.append(torch.tensor(out['bboxes']))
+...         categories.append(out['category'])
+...     return {'image': images, 'bbox': bboxes, 'category': categories}
+```
+
+Use the [`~Dataset.set_transform`] function to apply the transform on-the-fly which consumes less disk space. The randomness of data augmentation may return a different image if you access the same example twice. It is especially useful when training a model for several epochs.
+
+```py
+>>> ds['train'].set_transform(transforms)
+```
+
+You can verify the transform works by visualizing the 10th example:
+
+```py
+>>> example = ds['train'][10]
+>>> to_pil_image(
+...     draw_bounding_boxes(
+...         example['image'],
+...         box_convert(example['bbox'], 'xywh', 'xyxy'),
+...         colors='red',
+...         labels=[categories.int2str(x) for x in example['category']]
+...     )
+... )
+```
+
+<div class="flex justify-center">
+    <img src="https://huggingface.co/datasets/nateraw/documentation-images/resolve/main/visualize_detection_example_transformed_2.png">
+</div>

docs/source/image_process.mdx

@@ -151,7 +151,7 @@
     "bert_score>=0.3.6",
     "jiwer",
     "mauve-text",
-    "rouge_score",
+    "rouge_score<0.0.7",
     "sacrebleu",
     "sacremoses",
     "scikit-learn",

setup.py

@@ -27,7 +27,6 @@
 from collections import Counter, UserDict
 from collections.abc import Mapping
 from copy import deepcopy
-from dataclasses import asdict
 from functools import partial, wraps
 from io import BytesIO
 from math import ceil, floor
@@ -95,7 +94,7 @@
 from .utils._hf_hub_fixes import create_repo
 from .utils.file_utils import _retry, cached_path, estimate_dataset_size, hf_hub_url
 from .utils.info_utils import is_small_dataset
-from .utils.py_utils import convert_file_size_to_int, unique_values
+from .utils.py_utils import asdict, convert_file_size_to_int, unique_values
 from .utils.stratify import stratified_shuffle_split_generate_indices
 from .utils.tf_utils import minimal_tf_collate_fn
 from .utils.typing import PathLike
@@ -825,8 +824,13 @@ def from_pandas(
             info = DatasetInfo()
         info.features = features
         table = InMemoryTable.from_pandas(
-            df=df, preserve_index=preserve_index, schema=features.arrow_schema if features is not None else None
+            df=df,
+            preserve_index=preserve_index,
         )
+        if features is not None:
+            # more expensive cast than InMemoryTable.from_pandas(..., schema=features.arrow_schema)
+            # needed to support the str to Audio conversion for instance
+            table = table.cast(features.arrow_schema)
         return cls(table, info=info, split=split)
 
     @classmethod

src/datasets/arrow_dataset.py

@@ -16,7 +16,6 @@
 import json
 import os
 import sys
-from dataclasses import asdict
 from typing import Any, Dict, Iterable, List, Optional, Tuple, Union
 
 import fsspec
@@ -42,7 +41,7 @@
 from .table import array_cast, cast_array_to_feature, table_cast
 from .utils import logging
 from .utils.file_utils import hash_url_to_filename
-from .utils.py_utils import first_non_null_value
+from .utils.py_utils import asdict, first_non_null_value
 
 
 logger = logging.get_logger(__name__)