Datasets:
Tasks:
Image-Text-to-Text
Modalities:
Text
Formats:
parquet
Size:
1M - 10M
Tags:
image-text
visual-question-answering
image-captioning
instruction-tuning
pretraining
tinyvision
License:
conversations stringlengths 126 2.96k | image_bytes unknown |
|---|---|
[{"role": "user", "content": "Could you please interpret the image and write a detailed caption?<image>"}, {"role": "assistant", "content": "The image showcases a coastal scene with a prominent, well-defined wave in the foreground, curling into a classic tube shape that is characteristic of a good surfing wave. The wav... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "请用文字描绘这幅图片。<image>"}, {"role": "assistant", "content": "这幅图片捕捉到了一个原始的海滩场景，重点是正在破裂的一大波浪。这个波浪有一个明确定义的管状形状，展示了一个典型的中空气袋形态，表明这是冲浪的理想之选。海水呈现出深蓝绿色，表明水体干净并可能很深。在背景中，一座山脉高耸，部分被轻微的薄雾笼罩，这可能暗示着雾气腾腾的气候或是拍摄位置远离。前景的海滩沙质，看上去荒凉，看不到任何人类活动或野生动物的迹象。阳光照射在波浪的喷溅上，增强了水的肌理和动态。整个画面的情绪是自然之美和宁静。", "reasoning_content... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "<image>\n\n请用文字描述这幅图像。"}, {"role": "assistant", "content": "这幅图像捕捉到了一幅原始海滩场景，重点是一大波浪正在破浪的过程中。这个波浪有一个明显的管状形状，展现出一个带有空气口的桶形，暗示适合冲浪。海水呈深蓝绿色，表明水体清洁且可能很深。背景中，山脉若隐若现，部分被轻微的薄雾笼罩，可能暗示着多雾的气候或者照片拍摄位置的远处。前景的沙滩沙质细软，看起来空无一人，没有人类活动或野生动物的迹象。阳光洒在波浪飞溅的水花上，增强了水的质感和运动感。整幅图像的氛围是自然美和宁静。", "reasoning_content": ... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "<image>\nPlease enlighten me with a detailed description of the image."}, {"role": "assistant", "content": "The image depicts a modern hatchback car parked on a paved area with a scenic backdrop of rugged mountains under a soft twilight sky. The car is positioned in a three-quarter view sh... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "<image> 可以将图像中的视觉元素转化为文字吗？"}, {"role": "assistant", "content": "这幅图像呈现了一辆停放在柏油路上的深色掀背车，背景是一座山脉，时间看起来是黎明或黄昏，因为柔和的光线和长长的影子。车子呈现出后三分之一角度，展示其右侧。它采用了时尚的设计，一道明显的车身线从后车门延伸至尾灯，还配备了五辐合金轮毂。车辆的后部标识表明它是大众品牌的一款“GOLF”车型，额外的字母表明它是“TDI”变种，这意味着它配备了涡轮增压直喷柴油发动机。车牌可见，但文字太小无法清晰阅读。周围的景观崎岖险峻，天空大多晴朗，偶尔飘着一些薄云，营造出宁... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "你能将图片中的视觉元素转化为文字吗？ <image>"}, {"role": "assistant", "content": "这幅图片呈现了一辆停在沥青路上的深色掀背车，背景是在黎明或黄昏时分的山脉，柔和的光线和长长的影子似乎表明了这一点。汽车以一个后方三分之一的角度停放，展示了它的右侧。它设计时尚，有一条从后门延伸到尾灯的显著车身线，并配有五辐合金轮毂。车辆的后标识显示它是“大众”品牌的一款车型，“GOLF”，额外的字母表明它是“TDI”版本，这意味着它采用了涡轮增压直喷柴油引擎。车牌可见，但字体太小无法清晰阅读。周围的景观崎岖多峰，天空大部分晴朗，有几朵轻柔的云，营造... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "<image>\nPlease enlighten me with a detailed description of the image."}, {"role": "assistant", "content": "The image depicts a modern architectural residence with a clear sky in the background. The structure features a bold geometric design with rectangular shapes and a cantilevered upper... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
254,
0,
59,
67,
82,
69,
65,
84,
79,
82,
58,
32,
103,
100,
45,
106,
112,
101,
103,
32,
118,
49,
46,
48,
32,
40,
117,
115,
105,
110,
103,
... |
[{"role": "user", "content": "<image>\n\n是什么？请用您的话描述这张图片。"}, {"role": "assistant", "content": "这张图片显示了一个现代建筑结构，可能是一座住宅建筑。建筑物呈几何设计，拥有锐利的角度和干净的线条。其中有一个突出的悬挑部分向外延伸，大面积的玻璃窗让人可以看到内部。建筑主要由混凝土和玻璃材料构成，部分区域采用有纹理的瓷砖外墙。楼梯通往入口处，两侧有高架花园花坛。这些花坛种植了精心维护的绿植和多肉植物。天空晴朗蓝天，表明是一个阳光明媚的好天气。建筑的整体美观简约现代。", "reasoning_content": ""}] | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
254,
0,
59,
67,
82,
69,
65,
84,
79,
82,
58,
32,
103,
100,
45,
106,
112,
101,
103,
32,
118,
49,
46,
48,
32,
40,
117,
115,
105,
110,
103,
... |
[{"role": "user", "content": "<image>是什么样的图片？请用您的话描述一下这幅图片。"}, {"role": "assistant", "content": "这幅图片展示了一个现代化的建筑结构，可能是一栋住宅建筑。这座建筑具有几何设计，带有锐利的角度和整洁的线条。它拥有一个突出的悬挑部分，向外延伸，大面积的玻璃窗使内部一览无余。这个结构由混凝土和玻璃材料构成，一些部分还有带有纹理的瓷砖外观。一条室外楼梯通向入口，两侧有高起的花园花坛。这些花坛种植了精心维护的绿植和多肉植物。天空晴朗蔚蓝，表明是一个阳光明媚的好天气。整体建筑美学简约现代。", "reasoning_content": ""}] | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
254,
0,
59,
67,
82,
69,
65,
84,
79,
82,
58,
32,
103,
100,
45,
106,
112,
101,
103,
32,
118,
49,
46,
48,
32,
40,
117,
115,
105,
110,
103,
... |
[{"role": "user", "content": "Please provide a detailed written representation of the image.\n\n<image>"}, {"role": "assistant", "content": "This image captures a serene rural landscape at what appears to be either dawn or dusk, due to the soft lighting. A vast field of lavender, showing a rich purple hue, stretches ac... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "请提供关于这幅图片的详细书面描述。<image>"}, {"role": "assistant", "content": "这幅图片捕捉到了一个宁静的乡村风景，看起来可能是黎明或黄昏，因为光线柔和。一片广阔的薰衣草田铺展在前景，呈现出丰富的紫色色调，并与中间的传统石头建筑群相连。这些建筑是古老的，墙壁已经风化，看起来像是农舍，配有百叶窗和多层结构，暗示着用途是居住。复杂的后面，景色逐渐延伸至一个翠绿、点缀着树木的山坡，在一个从地平线处温暖过渡到上方凉爽的晴朗天空下。一条土路穿过薰衣草田，通往农舍。画面中没有人或现代机械，给人一种永恒的质感。", "reasoning_con... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "Could you please transcribe the image into a descriptive paragraph?\n\n<image>"}, {"role": "assistant", "content": "The image shows a portable, rectangular first aid kit with a textile exterior. It's an olive green color, consistent with military or outdoor utility gear. The kit is vertica... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "你能帮忙将这张图片转录成一个描述性段落吗？ <image>"}, {"role": "assistant", "content": "这张图片展示了一个带有纺织外观的便携式长方形急救包。它是橄榄绿色的，与军用或户外实用装备一致。急救包是垂直方向的，看起来是关闭状态，顶部的翻盖覆盖在前面。在翻盖上，有一个醒目的白色十字架，这在全球都被认为是医疗救助的象征。十字架下方，用粗体大写的无衬线字体印有“急救包”字样，颜色略轻于面料的颜色，确保易读性。在翻盖的右下角，有一个四角圆角的小方形标签。该标签看起来很暗，可能是黑色，上面印有白色或浅色的字母，其中“MFH”位于顶部，“INT. ... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "<image>\nPlease depict the image in words."}, {"role": "assistant", "content": "The image displays a text logo with a stylized effect. The main text reads \"Master of None\" which is the focal point of the design. The font is bold and has a modern, sans-serif typeface. The letters are whit... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "描述 <image>"}, {"role": "assistant", "content": "这幅图显示了一个带有样式效果的文本标志。主要文字是“无我之师”，这是设计的焦点。字体粗细，采用现代风格的无衬线字体。字母是白色的，带有三重轮廓效果：最内层是品红色，然后是绿色，再然后是一层薄薄的紫色。这创造了一种类似霓虹灯般的三维外观。在主要文本下方，有一段较小的文本，以标准无衬线字体写着“AfterBuzz TV AfterShow”，主要是白色，带有微妙的品红色轮廓。图片的背景是纯黑色，与文本的鲜艳颜色形成强烈对比，增强了霓虹灯效果。这幅图像除了文本之外，没有包含任何可辨认的... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "<image>Could you please express the image in a descriptive format?"}, {"role": "assistant", "content": "The image displays a vertical poster with a floral background. The flowers are white with some purple and green foliage, giving off a serene, natural feel. Superimposed on this floral ba... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "您能够描述一下这张图片吗？ <image>"}, {"role": "assistant", "content": "这张图片展示了一个有花朵背景的垂直海报。花朵是白色的，带有一些紫色和绿色的叶子，散发出一种宁静自然的感觉。在这个花卉背景的前景中，有一个白色、稍微透明的方框，里面包含有文本。文本采用粗体、无衬线字体书写，中心信息最为突出。文本中写着“THE MOST IMPORTANT LESSONS LEARNED FRESHMAN YEAR”（大一最重要的教训）。在这个主要信息下方，用更小的字体写着一个网址：“WWW.ENDLESSMAY.COM”。整体图片的美学风格是... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
219,
0,
67,
0,
20,
14,
15,
18,
15,
13,
20,
18,
16,
18,
23,
21,
20,
24,
30,
50,
33,
30,
28,
28,
30,
61,
44,
46,
36,
50,
73,
64,
76,
... |
[{"role": "user", "content": "Write a comprehensive caption for the image provided.\n<image>"}, {"role": "assistant", "content": "This panoramic image captures a serene mountainous landscape with a vibrant and diverse ecosystem. The foreground features crystal-clear waters where aquatic plants can be seen beneath the s... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
254,
0,
59,
67,
82,
69,
65,
84,
79,
82,
58,
32,
103,
100,
45,
106,
112,
101,
103,
32,
118,
49,
46,
48,
32,
40,
117,
115,
105,
110,
103,
... |
[{"role": "user", "content": "的综合描述<image>"}, {"role": "assistant", "content": "这幅全景图捕捉到了一幅宁静的山区风景，拥有丰富多样的生态系统。前景展示了清澈见底的水域，水下可以看到水生植物。水的清澈度显露了底部光滑的石头，这可能是一条河流或湖泊。有一条带有红橙色色调的鱼，可能是一条鲑鱼，正在跃出水面，半空中飞跃，周围可见水滴。中景主要由岩石形成的瀑布组成，形成多道瀑布，形成了增添氛围的薄雾。各种植被分布在地形上，包括针叶树和较小的灌木。在远景中，山峰拔地而起，背景是蓝天和零星的云。这片风景看起来野性而未经触摸，暗示着一个自然公园或偏远的荒野地区。", "... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
254,
0,
59,
67,
82,
69,
65,
84,
79,
82,
58,
32,
103,
100,
45,
106,
112,
101,
103,
32,
118,
49,
46,
48,
32,
40,
117,
115,
105,
110,
103,
... |
[{"role": "user", "content": "Please illustrate the image through your words.\n<image>"}, {"role": "assistant", "content": "The image features a promotional graphic for what appears to be a music-related event. The backdrop shows a dramatic sky at dusk or dawn, with vivid orange and purple hues. In front of this backdr... | [
255,
216,
255,
224,
0,
16,
74,
70,
73,
70,
0,
1,
1,
0,
0,
1,
0,
1,
0,
0,
255,
254,
0,
59,
67,
82,
69,
65,
84,
79,
82,
58,
32,
103,
100,
45,
106,
112,
101,
103,
32,
118,
49,
46,
48,
32,
40,
117,
115,
105,
110,
103,
... |
End of preview. Expand in Data Studio
TinyVision VLM Data
Training data for TinyVision, a small multimodal (image → text) model.
Both splits are 256×256 image–text pairs stored as sharded Parquet files (41 + 47 shards),
with the raw JPEG bytes embedded in the image_bytes column.
| Config | File | Rows | Size | Stage |
|---|---|---|---|---|
pretrain_i2t |
pretrain_i2t/part-*.parquet (41 shards) |
1,274,698 | 4.53 GB | Image–text pretraining (captioning / description) |
sft_i2t |
sft_i2t/part-*.parquet (47 shards) |
2,904,511 | 5.30 GB | Supervised fine-tuning (visual question answering) |
Schema
| Column | Type | Description |
|---|---|---|
conversations |
string | JSON-encoded message list, single-turn user → assistant |
image_bytes |
binary | Raw JPEG bytes, all images resized to 256×256 |
conversations example:
[
{"role": "user", "content": "Could you please interpret the image and write a detailed caption?<image>"},
{"role": "assistant", "content": "The image showcases a coastal scene ..."}
]
The <image> placeholder marks where the visual token is inserted.
Its position varies between samples (before or after the question text),
so the loader must handle it as an in-text marker rather than a fixed prefix.
Statistics
- Images: 100% 256×256, average 6–7 KB per JPEG (sampled)
- Turns: exactly 2 per sample (
user+assistant) - Language: about 46% of samples contain Chinese, the rest are English
- Sharding: 41 files (
pretrain_i2t) / 47 files (sft_i2t), each a standalone Parquet; streaming works per shard
License
Released under CC BY 4.0.
Please credit FLCgeigeigei/tinyvision-vlm-data when using this data.
- Downloads last month
- 130