Datasets:
input_png unknown | target_png unknown | caption stringclasses 95
values | source stringclasses 4
values |
|---|---|---|---|
[
137,
80,
78,
71,
13,
10,
26,
10,
0,
0,
0,
13,
73,
72,
68,
82,
0,
0,
2,
0,
0,
0,
2,
0,
8,
2,
0,
0,
0,
123,
26,
67,
173,
0,
0,
208,
15,
73,
68,
65,
84,
120,
156,
236,
253,
121,
184,
109,
215,
85,
222,
9,
191,
99,... | [
137,
80,
78,
71,
13,
10,
26,
10,
0,
0,
0,
13,
73,
72,
68,
82,
0,
0,
2,
0,
0,
0,
2,
0,
8,
2,
0,
0,
0,
123,
26,
67,
173,
0,
0,
187,
239,
73,
68,
65,
84,
120,
156,
236,
189,
119,
160,
100,
215,
81,
38,
94,
85,
231... | <mva> back-left view, high-angle shot | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAADQD0lEQVR4nOz9ebht11XeCb9jzLnWbk5777mNWku2ZWxjLLeyJFt(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nOz9d7gkR3YfiJ4Tkaa8u972bYtueAxmBsQYcih60Yj(...TRUNCATED) | <mva> right side view, top-down shot close-up | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAABP+ElEQVR4nO3de/CtWVoX9u/zPOvde/8u53T36e7TPZeeGeYCM3J(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAClsklEQVR4nOz96c91y3UfiP3Wqtr7DM/wDncmKZKiJIstyZJjyVb(...TRUNCATED) | <mva> front-right quarter view, eye-level shot close-up | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAABP+ElEQVR4nO3de/CtWVoX9u/zPOvde/8u53T36e7TPZeeGeYCM3J(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nOz9d7MkyXUnCp5zPETqK0urlgBakABBglrMknyzT9j(...TRUNCATED) | <mva> back-right view, top-down shot close-up | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nOz9aaBlW1IWin4RY8611t47m5Onq46iKLqCgleggk/(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nOz9d9AtSXYfBp5zMstc+9nnTXvfPd43BmYGAAEQwII(...TRUNCATED) | <mva> right side view, elevated shot close-up | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nOz9aaBlW1IWin4RY8611t47m5Onq46iKLqCgleggk/(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAEAAElEQVR4nOy9d4BkV3EuXlXn3Ntp8uagVU4EgUBCBJORAJtnkgG(...TRUNCATED) | <mva> back-left view, elevated shot | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAADY1UlEQVR4nOy9d6AkV3Xnf865t6q6+8UJ0mgkBBJhF3vXa+9v92e(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAAD3H0lEQVR4nOy9ebxuV1HnXVVr7f3MZz53HpKbEAJIAoEEwmRr46y(...TRUNCATED) | <mva> back view, elevated shot | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAADY1UlEQVR4nOy9d6AkV3Xnf865t6q6+8UJ0mgkBBJhF3vXa+9v92e(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAACxzElEQVR4nOz9a6zl25rehT3P+47xn3Outeqyr+fSfbppY3c3xna(...TRUNCATED) | <mva> right side view, top-down shot | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAACGDElEQVR4nOz9e7BnWXYX+H3XWnvvc87vcR/5qKyu6rcQEhIIiYE(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAADzSUlEQVR4nOz9a7B1W3oX9v2fyxhzrrX3+77nnG51S0IgyzK2cVU(...TRUNCATED) | <mva> back-right quarter view, high-angle shot | dome |
"iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAACGDElEQVR4nOz9e7BnWXYX+H3XWnvvc87vcR/5qKyu6rcQEhIIiYE(...TRUNCATED) | "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIACAIAAAB7GkOtAADbwElEQVR4nOz9aZBt2XXfB/7XsPc5997MfO/VBIKTe1LYHdGf+kO(...TRUNCATED) | <mva> front-left view, top-down shot | dome |
Qwen-Image-2.1 Multiple-Angles training pairs — v2
16,670 image-edit pairs in 18 parquet shards
(columns: input_png, target_png, caption, source).
v2 expands v1 (5,028 pairs, 2,514 objects) on the axis v1 was weakest: subject diversity, especially characters and creatures.
| Source | Shards | Objects | Pairs | License |
|---|---|---|---|---|
| zeyuanyin/Dome-Objaverse 48-view dome renders, category-filtered toward characters/creatures via Cap3D captions | train-0000{0-9} |
4,688 | 9,376 | renders CC-BY-4.0 (source meshes carry their own licenses — see Dome-Objaverse README) |
| Linzhan/Objaverse-XL-Rigged-Animated-Renders rest-pose grids of rigged characters/creatures, filtered to permissive upstream licenses (CC-BY / CC-BY-SA / CC0 / public domain only, verified per-asset via allenai/objaverse metadata) | train-0002{0-3} |
~1,986 | 3,952 | per-asset, filtered permissive |
| Dome objects composited on real Unsplash backgrounds + background-reconstruction pairs (v2.1 lever: the synthetic-studio domain gap) | train-0003{0-3} |
1,114 | 3,342 | backgrounds: Unsplash license |
The v2 LoRA trained on the first two sources (13,328 pairs) — the real-background
shards landed while training was underway and are reserved for a future v2.1. See
manifest.json for the exact per-source layout and training-subset mapping.
Caption format (unchanged from v1, trigger <mva>):
<mva> {azimuth}, {elevation}[ close-up] — 12 azimuths × 30°, elevations
eye-level/elevated/high-angle/top-down, close-up = 62% center crop.
Input = front view composited on a light-grey studio background (RGB 211,211,211); targets = other views at labeled camera poses. Linzhan pairs are eye-level front → back / right-side views from the rest-pose grids (front/back on the top row, left/right below).
Rejects: 1,512 dome objects (quality filters), ~5,387 Linzhan assets (non-permissive or unverifiable license), grids failing variance checks.
- Downloads last month
- 252