WesleyHsieh0806 commited on
Commit
d5c4b1d
β€’
1 Parent(s): 0be95e3
Files changed (1) hide show
  1. README.md +148 -0
README.md ADDED
@@ -0,0 +1,148 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # TAO-Amodal Dataset
2
+
3
+ <!-- Provide a quick summary of the dataset. -->
4
+ Official Source for Downloading the TAO-Amodal Dataset.
5
+
6
+ [**πŸ“™ Project Page**](https://tao-amodal.github.io/) | [**πŸ’» Code**](https://github.com/WesleyHsieh0806/TAO-Amodal) | [**πŸ“Ž Paper Link**](https://arxiv.org/abs/2312.12433) | [**✏️ Citations**](#citations)
7
+
8
+ <div align="center">
9
+ <a href="https://tao-amodal.github.io/"><img width="95%" alt="TAO-Amodal" src="https://tao-amodal.github.io/static/images/webpage_preview.png"></a>
10
+ </div>
11
+
12
+ </br>
13
+
14
+ Contact: [πŸ™‹πŸ»β€β™‚οΈCheng-Yen (Wesley) Hsieh](https://wesleyhsieh0806.github.io/)
15
+
16
+ ## Dataset Description
17
+ Our dataset augments the TAO dataset with amodal bounding box annotations for fully invisible, out-of-frame, and occluded objects.
18
+ Note that this implies TAO-Amodal also includes modal segmentation masks (as visualized in the color overlays above).
19
+ Our dataset encompasses 880 categories, aimed at assessing the occlusion reasoning capabilities of current trackers
20
+ through the paradigm of Tracking Any Object with Amodal perception (TAO-Amodal).
21
+
22
+ ### Dataset Download
23
+ 1. Download all the annotations.
24
+ ```bash
25
+ git lfs install
26
+ git clone git@hf.co:datasets/chengyenhsieh/TAO-Amodal
27
+ ```
28
+
29
+ 2. Download all the video frames:
30
+
31
+ You can either download the frames following the instructions [here](https://motchallenge.net/tao_download.php) (recommended) or modify our provided [script](./download_TAO.sh) and run
32
+ ```bash
33
+ bash download_TAO.sh
34
+ ```
35
+
36
+
37
+
38
+
39
+ ## πŸ“š Dataset Structure
40
+
41
+ The dataset should be structured like this:
42
+ ```bash
43
+ β”œβ”€β”€ frames
44
+ └── train
45
+ β”œβ”€β”€ ArgoVerse
46
+ β”œβ”€β”€ BDD
47
+ β”œβ”€β”€ Charades
48
+ β”œβ”€β”€ HACS
49
+ β”œβ”€β”€ LaSOT
50
+ └── YFCC100M
51
+ β”œβ”€β”€ amodal_annotations
52
+ β”œβ”€β”€ train/validation/test.json
53
+ β”œβ”€β”€ train_lvis_v1.json
54
+ └── validation_lvis_v1.json
55
+ β”œβ”€β”€ example_output
56
+ └── prediction.json
57
+ └── BURST_annotations
58
+ └── train
59
+ └── train_visibility.json
60
+
61
+ ```
62
+
63
+ ## πŸ“š File Descriptions
64
+
65
+ | File Name | Description |
66
+ | -------------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
67
+ | train/validation/test.json | Formal annotation files. We use these annotations for visualization. Categories include those in [lvis](https://www.lvisdataset.org/) v0.5 and freeform categories. |
68
+ | train_lvis_v1.json | We use this file to train our [amodal-expander](https://tao-amodal.github.io/index.html#Amodal-Expander), treating each image frame as an independent sequence. Categories are aligned with those in lvis v1.0. |
69
+ | validation_lvis_v1.json | We use this file to evaluate our [amodal-expander](https://tao-amodal.github.io/index.html#Amodal-Expander). Categories are aligned with those in lvis v1.0. |
70
+ | prediction.json | Example output json from amodal-expander. Tracker predictions should be structured like this file to be evaluated with our [evaluation toolkit](https://github.com/WesleyHsieh0806/TAO-Amodal?tab=readme-ov-file#bar_chart-evaluation). |
71
+ | BURST_annotations/XXX.json | Modal mask annotations from [BURST dataset](https://github.com/Ali2500/BURST-benchmark) with our heuristic visibility attributes. We provide these files for the convenience of visualization |
72
+
73
+ ### Annotation and Prediction Format
74
+
75
+ Our annotations are structured similarly as [TAO](https://github.com/TAO-Dataset/annotations) with some modifications.
76
+ Annotations:
77
+ ```bash
78
+
79
+ Annotation file format:
80
+ {
81
+ "info" : info,
82
+ "images" : [image],
83
+ "videos": [video],
84
+ "tracks": [track],
85
+ "annotations" : [annotation],
86
+ "categories": [category],
87
+ "licenses" : [license],
88
+ }
89
+ annotation: {
90
+ "id": int,
91
+ "image_id": int,
92
+ "track_id": int,
93
+ "bbox": [x,y,width,height],
94
+ "area": float,
95
+
96
+ # Redundant field for compatibility with COCO scripts
97
+ "category_id": int,
98
+ "video_id": int,
99
+
100
+ # Other important attributes for evaluation on TAO-Amodal
101
+ "amodal_bbox": [x,y,width,height],
102
+ "amodal_is_uncertain": bool,
103
+ "visibility": float, (0.~1.0)
104
+ }
105
+ image, info, video, track, category, licenses, : Same as TAO
106
+ ```
107
+
108
+ Predictions should be structured as:
109
+
110
+ ```bash
111
+ [{
112
+ "image_id" : int,
113
+ "category_id" : int,
114
+ "bbox" : [x,y,width,height],
115
+ "score" : float,
116
+ "track_id": int,
117
+ "video_id": int
118
+ }]
119
+ ```
120
+ Refer to the instructions of [TAO dataset](https://github.com/TAO-Dataset/tao/blob/master/docs/evaluation.md) for further details
121
+
122
+
123
+ ## πŸ“Ί Example Sequences
124
+ Check [here](https://tao-amodal.github.io/#TAO-Amodal) for more examples and [here](https://github.com/WesleyHsieh0806/TAO-Amodal?tab=readme-ov-file#artist-visualization) for visualization code.
125
+ [<img src="https://tao-amodal.github.io/static/images/car_and_bus.png" width="50%">](https://tao-amodal.github.io/dataset.html "tao-amodal")
126
+
127
+
128
+
129
+ ## Citation
130
+
131
+ <!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
132
+ ```
133
+ @misc{hsieh2023tracking,
134
+ title={Tracking Any Object Amodally},
135
+ author={Cheng-Yen Hsieh and Tarasha Khurana and Achal Dave and Deva Ramanan},
136
+ year={2023},
137
+ eprint={2312.12433},
138
+ archivePrefix={arXiv},
139
+ primaryClass={cs.CV}
140
+ }
141
+ ```
142
+
143
+ ---
144
+ task_categories:
145
+ - object-detection
146
+ - multi-object-tracking
147
+ license: mit
148
+ ---