InstVL: A Large-Scale Instance-Aware Vision-Language Dataset

This is the official repository for the InstVL dataset, introduced in the paper InstAP: Instance-Aware Vision-Language Pre-Train for Spatial-Temporal Understanding.

InstVL is a large-scale dataset of images and videos designed to bridge the gap between holistic scene understanding and fine-grained, instance-level comprehension. Current vision-language pre-training (VLP) paradigms excel at global scene understanding but struggle with instance-level reasoning due to global-only supervision. InstVL addresses this by providing dual-granularity annotations that enable instance-aware pre-training.

InstVL provides two levels of detailed textual annotations:

• Global Captions — A comprehensive description of the entire scene.
• Instance Captions — Fine-grained descriptions grounded to specific object regions (in images) or spatial-temporal trajectories (in videos).

The dataset contains 2 million images and 50,000 videos, providing rich supervision for instance-centric pre-training and benchmarking.

---

💻 How to Load with 🤗 Datasets

You can load the dataset directly using the Hugging Face datasets library. You must specify which configuration (image or video) you want to load.

Loading the Image Dataset

from datasets import load_dataset

img_ds = load_dataset(
    "wovenbytoyota-vai/instvl",
    name="image",
    data_files={
        "train": "train/instvl_img_2m.jsonl",
        "test_1k": "test/instvl_img_1k.jsonl",
        "test_10k": "test/instvl_img_10k.jsonl",
        "test_zero_1k": "test/instvl_img_zero_1k.jsonl",
        "test_zero_10k": "test/instvl_img_zero_10k.jsonl",
    },
)

# Access a split
train_split = img_ds["train"]

print(train_split[0])

Loading the Video Dataset

from datasets import load_dataset

video_ds = load_dataset(
    "wovenbytoyota-vai/instvl",
    name="video",
    data_files={
        "train": "train/instvl_video_50k.jsonl",
        "test_1k": "test/instvl_video_1k.jsonl",
        "test_10k": "test/instvl_video_10k.jsonl",
    }
)
# Access a split
train_split = video_ds["train"]

# Print the first example
print(train_split[0])

---

📋 Dataset Structure

The dataset is organized into train and test splits, with data provided in the JSON Lines (.jsonl) format.

.
├── test
│   ├── instvl_img_10k.jsonl
│   ├── instvl_img_1k.jsonl
│   ├── instvl_img_zero_10k.jsonl
│   ├── instvl_img_zero_1k.jsonl
│   ├── instvl_video_10k.jsonl
│   └── instvl_video_1k.jsonl
└── train
    ├── instvl_img_2m.jsonl
    └── instvl_video_50k.jsonl

---

🖼️ Image Data Structure

Each line in the image .jsonl files represents a single image and its annotations.

Fields

Key	Data Type	Description
image	String	The relative file path to the JPG image.
caption	String	A detailed, holistic caption describing the entire image scene.
image_id	String	A unique identifier for the image.
instance_data	List of Objects	A list of annotations for specific object instances within the image. Each annotation pairs a bounding box with a free-form sentence describing that instance. Can be [].

instance_data Object

Key	Data Type	Description
instance_id	String	A unique ID for the detected object within the image.
instance_category	String	The category or class assigned to the object (e.g., "Shoes", "Player").
bbox	List of Int	The bounding box coordinates for the object, in [x, y, width, height] format.
instance_caption	String	A fine-grained caption describing only the object within the bounding box.

---

📹 Video Data Structure

Each line in the video .jsonl files represents a single video segment and its annotations.

Fields

Key	Data Type	Description
video	String	The filename of the segmented MP4 video clip.
resolution	List of Int	The video's resolution as [height, width].
caption	String	A detailed summary of the events and scene within the entire video segment.
duration	Float	The total duration of the video segment in seconds.
segment_frame_range	List of Int	The [start_frame, end_frame] numbers from the original, full-length source video.
instance_data	List of Objects	A list containing information about each object tracked across multiple frames. The instance caption is shared by all boxes on an object's trajectory.

instance_data Object

Key	Data Type	Description
instance_id	Integer	A unique ID for the tracked object within this video segment.
instance_category	String	The category assigned to the tracked object (e.g., "person", "furniture").
instance_caption	String	A summary describing the object and its actions throughout the video.
frames	List of Obj	A list of frame-wise boxes for the object.
frame	Integer	(Field inside each frames item) The frame index.
bbox	List of Int	(Field inside each frames item) The bounding box coordinates [x, y, width, height] for the object on a specific frame.

---

📊 Data Splits

The InstVL dataset is divided into several training and test splits to facilitate robust benchmarking.

Split	Filename	# of Samples	Description
Train	train/instvl_img_2m.jsonl	~2,000,000	Main training set for images.
Train	train/instvl_video_50k.jsonl	~50,000	Main training set for videos.
Test	test/instvl_img_1k.jsonl	2,442	Standard image test set.
Test	test/instvl_img_10k.jsonl	23,129	Larger image test set.
Test	test/instvl_video_1k.jsonl	2,508	Standard video test set.
Test-Zero	test/instvl_img_zero_1k.jsonl	2,570	Zero-shot image test set.
Test-Zero	test/instvl_img_zero_10k.jsonl	26,029	Larger zero-shot image test set.

Meaning of the "Zero" Splits

The img-zero splits are a key component for evaluation. These subsets are sourced entirely from COYO, whereas the main training images (and their corresponding test splits) are from LAION. This introduces a distribution shift that lets us confirm that a model's performance demonstrates true generalization capabilities and is not merely inherited from the training distribution.

---

🏆 Benchmark Results

The following tables show the performance of the InstAP model on the InstVL test sets and standard zero-shot benchmarks, compared to other state-of-the-art models. All results are from the original paper. UMT-L (InstVL; g) and UMT-L (InstVL; g+i) are baselines trained on the same full training corpus as InstAP, using only InstVL's global captions (g) or with all InstVL captions treated as global (g+i).

Instance-Level Retrieval Performance (R@1)

This task evaluates the model's ability to retrieve the correct fine-grained instance caption. We report T2V / V2T R@1.

Method	InstVL(img) 1K	InstVL(img) 10K	InstVL(img-zero) 1K	InstVL(img-zero) 10K	InstVL(video) 1K
VideoPrism	28.21 / 34.52	22.75 / 29.51	21.32 / 27.39	13.85 / 20.04	40.86 / 39.29
CLIP4Clip	25.10 / 33.21	18.68 / 28.19	17.82 / 25.10	9.11 / 16.30	17.71 / 24.69
Coca	11.83 / 21.79	7.36 / 13.33	7.08 / 13.19	4.12 / 7.26	14.72 / 11.82
ViCLIP	28.38 / 28.91	19.46 / 20.02	18.25 / 20.93	9.57 / 11.21	21.78 / 21.50
OpenCLIP	37.88 / 44.06	29.21 / 37.76	26.73 / 36.19	17.28 / 25.57	36.63 / 33.36
CLIP-ViP	24.04 / 32.06	14.38 / 21.85	13.81 / 22.96	6.60 / 12.11	16.78 / 28.32
MCQ	19.33 / 22.11	9.63 / 11.13	17.08 / 19.61	7.04 / 8.55	24.41 / 23.72
SigLIP	38.17 / 45.17	29.76 / 37.83	28.25 / 35.56	16.98 / 25.19	36.43 / 36.14
UMT-L	38.44 / 35.65	21.34 / 23.08	29.34 / 30.17	11.09 / 16.38	26.38 / 22.43
UMT-L (InstVL; g)	34.44 / 41.24	22.87 / 30.37	25.97 / 31.97	13.33 / 19.21	41.51 / 40.34
UMT-L (InstVL; g+i)	45.74 / 44.27	34.83 / 35.15	34.68 / 34.99	21.13 / 22.82	40.38 / 39.33
InstAP (Ours)	50.25 / 49.26	44.05 / 45.76	41.94 / 42.53	28.25 / 31.87	60.63 / 58.49

Global Retrieval Performance (R@1)

This task evaluates the model's ability to retrieve the correct global caption for the entire scene. We report T2V / V2T R@1.

Method	InstVL(img) 1K	InstVL(img) 10K	InstVL(img-zero) 1K	InstVL(img-zero) 10K	InstVL(video) 1K
VideoPrism	97.40 / 97.60	88.19 / 89.62	85.70 / 85.80	73.05 / 75.11	82.71 / 83.62
CLIP4Clip	93.40 / 96.00	79.22 / 84.25	78.20 / 81.70	56.95 / 63.96	67.50 / 70.50
Coca	86.20 / 91.50	70.80 / 76.16	67.40 / 70.50	46.05 / 50.64	46.92 / 43.78
ViCLIP	95.10 / 93.50	81.47 / 79.33	77.80 / 77.60	58.51 / 58.21	62.89 / 62.69
OpenCLIP	94.40 / 98.10	84.98 / 92.06	83.40 / 86.90	70.75 / 78.13	82.00 / 77.15
CLIP-ViP	78.40 / 89.20	54.94 / 72.00	55.60 / 73.20	32.48 / 51.30	35.59 / 61.07
MCQ	58.20 / 60.10	31.45 / 34.12	58.90 / 62.70	34.13 / 38.26	61.48 / 60.67
SigLIP	95.70 / 98.20	87.18 / 91.97	83.90 / 86.50	68.64 / 75.66	74.72 / 76.14
UMT-L	94.70 / 95.30	83.95 / 85.41	83.90 / 83.70	72.60 / 72.59	88.30 / 85.50
UMT-L (InstVL; g)	96.20 / 97.10	85.70 / 87.03	85.30 / 86.40	72.50 / 74.18	84.80 / 82.40
UMT-L (InstVL; g+i)	93.20 / 94.30	80.30 / 81.62	82.40 / 84.30	68.16 / 69.76	79.90 / 77.20
InstAP (Ours)	99.20 / 99.10	95.77 / 94.71	88.70 / 88.30	83.33 / 82.21	94.50 / 95.50

Zero-Shot Text-to-Video Retrieval (R@1 / R@5 / R@10)

This table evaluates generalization on standard zero-shot text-to-video retrieval benchmarks.

Method	MSR-VTT	DiDeMo	MSVD	LSMDC	ActivityNet
CLIP4Clip	32.0 / 57.0 / 66.9	--	38.5 / 66.9 / 76.8	15.1 / 28.5 / 36.4	--
Frozen in Time	18.7 / 39.5 / 51.6	21.1 / 46.0 / 56.2	38.7 / 70.1 / 80.1	9.3 / 22.0 / 30.1	--
VIOLET	25.9 / 49.5 / 59.7	23.5 / 49.8 / 59.8	--	--	--
ALPRO	24.1 / 44.7 / 55.4	23.8 / 47.3 / 57.9	--	--	--
RAP	28.9 / 47.5 / 56.8	29.5 / 55.7 / 65.6	35.9 / 64.3 / 73.7	12.8 / 26.6 / 33.4	--
Clover	26.4 / 49.5 / 60.0	29.5 / 55.2 / 66.3	--	14.7 / 29.2 / 38.2	--
TW-BERT	26.4 / 50.1 / 59.6	28.4 / 52.9 / 64.5	--	14.2 / 30.4 / 36.0	--
Singularity	28.4 / 50.2 / 59.5	36.9 / 52.9 / 64.5	--	--	--
LaT	23.4 / 44.1 / 53.3	22.6 / 45.9 / 58.9	36.9 / 68.6 / 81.0	--	--
OA-Trans	23.4 / 47.5 / 55.6	23.5 / 50.4 / 59.8	--	--	--
MCQ	26.0 / 46.4 / 56.4	25.6 / 50.6 / 61.1	43.6 / 74.9 / 84.9	12.2 / 25.9 / 32.2	--
MILES	26.1 / 47.2 / 56.9	27.2 / 50.3 / 63.6	44.4 / 76.2 / 87.0	11.1 / 24.7 / 30.6	--
CLIP-ViP	31.7 / 51.2 / 63.2	24.6 / 50.7 / 59.7	--	12.5 / 26.1 / 33.3	--
EA-VTR	28.0 / 53.1 / 62.3	32.7 / 58.9 / 68.9	46.6 / 78.9 / 86.5	15.7 / 29.6 / 36.0	--
UMT-L	39.7 / 61.8 / 70.9	47.0 / 71.8 / 78.8	47.0 / 75.4 / 83.6	26.0 / 43.1 / 51.6	44.3 / 72.2 / 84.4
UMT-L (InstVL; g)	35.4 / 59.4 / 70.2	44.1 / 72.3 / 79.1	43.7 / 73.4 / 82.4	19.9 / 38.4 / 46.5	39.8 / 66.5 / 76.5
UMT-L (InstVL; g+i)	34.0 / 58.5 / 68.5	42.7 / 69.0 / 77.0	41.3 / 71.8 / 81.4	17.5 / 36.6 / 46.5	37.1 / 64.5 / 74.7
InstAP (Ours)	41.1 / 65.2 / 73.6	54.0 / 78.2 / 84.5	49.2 / 77.0 / 85.1	23.5 / 42.7 / 50.3	50.7 / 77.2 / 86.6

Visual Grounding Performance (IoU@{50, 70, 90}) on InstVL-1K

This task evaluates the model's ability to localize textual mentions to specific spatial-temporal regions.

Method	InstVL(img) IoU@50 / @70 / @90	InstVL(img-zero) IoU@50 / @70 / @90	InstVL(video) IoU@50 / @70 / @90
UMT-L	74.53 / 63.47 / 41.64	67.12 / 54.20 / 34.05	54.25 / 40.70 / 14.44
InstAP (Ours)	76.17 / 67.04 / 48.20	68.52 / 58.91 / 42.14	60.02 / 48.85 / 25.13

---

⬇️ Downloading the Original Images & Videos

You can download the original images and videos from the following websites:

• InstVL Image: LAION-400M via img2dataset
• InstVL Image Zero: COYO-700M via img2dataset
• InstVL Video: HD-VILA-100M on Hugging Face

---

🙏 Acknowledgements

This dataset is based on results obtained from a project, JPNP20017, subsidized by the New Energy and Industrial Technology Development Organization (NEDO).

📝 License

Refer to the license CC BY-NC-SA 4.0 for using our dataset.

---

📜 Citations

If you use this dataset in your research, please cite the original paper:

@misc{kumar2026instapinstanceawarevisionlanguagepretrain,
  title={InstAP: Instance-Aware Vision-Language Pre-Train for Spatial-Temporal Understanding},
  author={Ashutosh Kumar and Rajat Saini and Jingjing Pan and Mustafa Erdogan and Mingfang Zhang and Betty Le Dem and Norimasa Kobori and Quan Kong},
  year={2026},
  eprint={2604.08337},
  archivePrefix={arXiv},
  primaryClass={cs.CV},
  url={https://arxiv.org/abs/2604.08337},
}