id stringclasses 10
values | task stringclasses 1
value | input_images images listlengths 1 1 | reasoning_thought_0 stringclasses 10
values | reasoning_image_0 imagewidth (px) 342 512 | reasoning_thought_1 stringclasses 10
values |
|---|---|---|---|---|---|
eb8050b5a0778054ba678f218f7a82ba4f120feebfbdf95ec1729e694bb77877 | colorization | A grayscale view shows a railway turntable yard in strong daylight, with train tracks and coarse ballast stones in the foreground leading straight onto a metal turntable bridge at the center. The bridge has raised side beams, a rail line down the middle, and small rectangular placards near the front, while a wide circu... | With the colors back in place, the yard sits under bright midday sunlight beneath a clean pale-to-medium blue sky crossed by several overhead utility wires. Steel tracks run over reddish-brown ballast toward the center and onto a cool gray metal turntable bridge with riveted side beams, a central rail line, and small w... | ||
1281b19983b6d1efd4cfac77cb63e329d2997e1aca17cb077f1d9208d2a6099f | colorization | A compact modern kitchen lines the left side of a bright interior in a fully desaturated view, so the scene reads in silvers, grays, and blacks rather than natural color. Smooth walls and ceiling frame a narrow galley-like layout, with a wall sconce glowing on the far left and a long countertop running back beneath upp... | With the colors back in place, the space becomes a crisp contemporary apartment kitchen with smooth white walls, ceiling, countertop, lower cabinets, column, and the broad louvered pantry or closet door on the right. Pale honey wood flooring warms the bottom of the scene, while the upper cabinets shift into warm medium... | ||
18e6b979a58fb8808c7b61f0776f05a8a762c357f11cbfdc23d9b8a6afd674a3 | colorization | An overhead industrial work area is shown in grayscale, centered on a deep metal bin with slanted sides sitting on a tabletop. The bin’s interior looks worn, with scratches and a dark mottled band along the back wall, and inside it there are two main loose items: a small dumbbell-shaped piece near the lower left corner... | With the colors back in place, the workshop scene takes on a cool, subdued industrial palette. The central bin is steel gray with faint bluish undertones, its interior marked by charcoal scuffs and a near-black splatter band across the back wall. Near the lower left inside the bin sits a small orange plastic piece with... | ||
95f43f07a2dae3f41fe1597fe2877eef6ee9bcc00fd4a7edfb3dc5c8e8c1e268 | colorization | A tidy contemporary living room is seen in grayscale, centered on a low square coffee table set over a thick shag rug, with a second lower table surface extending in from the left. On the main tabletop sits a stack of large books with a broad hardcover on top, a glass jar candle near the front edge, a small frosted vot... | With the colors back in place, the room resolves into a bright modern living room with creamy white tables, cabinetry, radiator, and media console set against warm honey-brown wood flooring. The thick shag rug under the coffee table shows a blend of silvery, charcoal, and pale gray tones. On the main white table, a viv... | ||
37921ed3a447a9453dabb9a8a64488b901875bf5cdf06f04b5dc040f601fcfc5 | colorization | An overhead grayscale view shows a tight cluster of a dozen decorated cupcakes packed into a dark tray, with only slim black channels visible between them. Most of the frame is filled by the cupcake tops, each in a ridged paper liner, and the frosting appears in smooth swirls while other areas are rough with crumbly to... | With the colors back in place, the dozen cupcakes read as a bright, playful arrangement in a deep black rectangular tray, seen from above with only narrow dark gaps between them. Each cupcake sits in a brown ridged paper liner and is topped with soft swirls of pale sky-blue frosting, while many are partly covered with ... | ||
8f3e7ec3d950da421644e967f2e4707335a39a92e05ba6358a915caf997ff7e7 | colorization | A grayscale view shows a dim restaurant or banquet room built with rough brick walls, centered on a round table in the foreground where seven seated adults gather in conversation. Their faces are blurred, but their postures suggest they are reading menus, eating, or looking toward one another, with one person at the fr... | With the colors back in place, the room becomes a cozy, intimate restaurant scene in deep brown and charcoal tones, with rough dark brick walls and black shelving warmed by amber wall sconces and chandelier reflections in the three tall dark-framed mirrors. The round foreground table is covered in a crisp white tablecl... | ||
9fe2c150bdbad6c38d7193d260152549d1af74ecd4318114f397c9da59ddc1c5 | colorization | This is a grayscale indoor view of a partially open interior door at the right, angled toward the viewer so its lever handle and rectangular latch plate are prominent in the foreground. On the left, an off-white-looking wall with a slightly rough cinderblock texture holds a simple rectangular light switch plate with a ... | With the colors back in place, the scene resolves into a soft, evenly lit interior with off-white to warm-white walls, trim, door, and switch plate, the cinderblock texture on the left wall picked out by faint gray shadowing. The door on the right remains partly open toward the viewer, fitted with cool brushed-silver h... | ||
097ff64d9e67a8c0273aac3cb032bee986ef8f84a14211dff184899427dd2c66 | colorization | A small flock of sheep is spread across a grassy field in a desaturated view, with one large woolly sheep dominating the center foreground and facing straight toward the camera while another sheep just behind it lowers its head to graze. In the upper left, a partially cropped sheep also bends down into the grass, and f... | With the colors back in place, the pasture becomes a vivid fresh green field under soft, diffuse overcast light, giving the whole farm scene a calm, even look with almost no hard shadows. The large sheep in the center foreground stands out with dense curly wool that is creamy off-white to light beige, with weathered gr... | ||
1becc0e41b905e9d8c43d263486d6511d9c05b865e5edb19dd7f8b98fdd16453 | colorization | A grayscale sports scene is frozen at the moment of an aerial contest on an open field. Three players dominate the middle: one tall player near center left is leaping high with legs spread and an arm stretched toward a flying disc, another player at center right is also airborne with one knee lifted and an arm extended... | With the colors back in place, a game of ultimate frisbee is unfolding on a grassy field under bright midday light, with a pale blue-white sky and crisp dark shadows cast to the lower right. The ground is mostly medium green, interrupted by a worn tan-brown dirt patch in the left foreground and a few drier yellowish ar... | ||
61dbd445d7630bed9bcd89ca6743f0085000e07ff6b89bad2fe5adfaf27ff74e | colorization | This is a grayscale view of a cozy open kitchen and adjoining living area, with the scene arranged around a rectangular island slightly right of center and a spindle-back chair in the foreground facing it. A dark counter edge cuts across the lower right, while above the island a suspended rack holds several pots and pa... | With the colors back in place, the room becomes a warm, wood-accented kitchen and living space filled with honey and chestnut tones in the beams, floors, cabinetry, and the deck outside, all set against soft creamy white walls. The central island, slightly right of center, has a pale butcher-block beige top over a dark... |
Pretext-270K
Pretext-270K is the first-stage training data of ThinkMorph-SSL-Base. It holds 270,896 training traces and 1,600 validation traces. Each trace starts from a corrupted input image. A description follows, then the restored image, then a further description. There is no question and no answer.
The traces are grouped into seven tasks, one per corruption that the model undoes.
Tasks
| Task | Input image | Restored image | Train rows | Val rows | Of these, text only |
|---|---|---|---|---|---|
jigsaw |
shuffled tiles: one image cut into tiles, or separate tile images | reassembled image | 52,975 | 571 | 10,266 |
orthogonal_transform |
rotated or flipped image | upright image | 36,763 | 183 | — |
deformation_recovery |
geometrically warped image | unwarped image | 36,672 | 189 | — |
super_resolution |
low-resolution image | high-resolution image | 36,409 | 179 | — |
colorization |
grayscale image | colored image | 36,104 | 168 | — |
denoising |
noisy image | clean image | 36,023 | 155 | — |
masked_inpainting |
image with masked regions | completed image | 35,950 | 155 | — |
| Total | 270,896 | 1,600 | 10,266 |
Stage 1 trained on the train rows only. Choose a task in the Dataset Viewer to see examples.
Files
| Path | Content |
|---|---|
shards/annotations.tar |
MANIFEST.json and every task file: all rows, no images |
shards/data-*.tar |
the 533,366 images (141 GB) in 133 tar shards |
shards/SHARDS.json |
size and SHA256 of every shard |
text_only/ |
10,266 rows whose images (Matterport3D) are not distributed here; see Rows without images |
preview/ |
the Dataset Viewer examples, one file per task |
Download
hf download ThinkMorph/Pretext-270K --repo-type dataset --include "shards/*" --local-dir Pretext-270K
mkdir -p data
for f in Pretext-270K/shards/*.tar; do tar -xf "$f" -C data; done
shards/SHARDS.jsongives each shard's SHA256, andMANIFEST.jsoneach extracted file's.- To read the rows without the images, download and extract
shards/annotations.taronly.
Layout and row format
After extracting the shards:
data/
MANIFEST.json
images/<sha256>.<ext> shared by all tasks
colorization/train.jsonl
colorization/val.jsonl
... one folder per task
{"id": "<sha256>", "task": "colorization", "images": ["../images/<sha256>.jpg"],
"reasoning_thought_0": "...", "reasoning_image_0": "../images/<sha256>.jpg",
"reasoning_thought_1": "..."}
imagesis the corrupted input.reasoning_image_0is the restored image.- Image paths are relative to the row's file.
idis a stable public id.
Rows without images
text_only/ holds the rows of the jigsaw task whose images come from Matterport3D. It has the same task folders as the main release.
The terms of their image sources do not allow us to redistribute the images.
text_only/images.jsonlnames every image. For each one it gives the upstream identifiers, how the image was derived, and the size and MD5 of the file used in training.- Put the images you obtain in
text_only/images/; the rows refer to them as../images/<name>.
To obtain the source data:
- Matterport3D: sign the Matterport3D Terms of Use. The form is at https://niessner.github.io/Matterport/; send it to matterport3d@googlegroups.com.
License
The images come from several public datasets, and each keeps its own terms. There is no blanket license for the images. A list of the image sources per task will be added with the paper.
- Downloads last month
- 105