Datasets:

ThinkMorph
/

Instruct-180K

id string	task string	question string	answer string	reasoning_thought_0 string	reasoning_thought_1 string	reasoning_thought_2 string	reasoning_image_2 image	reasoning_thought_3 string
aa45b3b692ee0e68c1c21b465859c1597edc802997818e57686a3c6517cd72af	birds_eye_view_rotation	Based on these four images (image 1, 2, 3, and 4) showing the powder bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: From the viewpoint presented in image 3, what is to the left of the powder bottle? A. Dining area B. Black cabinet C. Red bed frame wi...	A	I first inspect the queried view, the one with the red bed frame behind the powder bottle. The question asks what is to the left of the bottle from that specific viewpoint, so I need to convert this egocentric left/right relation into the room layout. To do that, I will generate a bird's-eye view of the tabletop and ne...	The first generated bird's-eye view shows the powder bottle near the center-right of the blue mat, with the black round container and green packet on one side, and the tissue pack and green rectangular pouch on the other. This top-down layout clarifies where objects sit around the bottle independent of perspective dist...	In the rotated bird's-eye view, the bottle's aligned forward direction matches the original viewpoint that looks out toward the open room with the table and chairs in the distance. That is the same orientation as the image showing the dining area beyond the tabletop. In that matched original view, the space directly ah...	Not supported with pagination yet
9e9b3ae7f7a309a92479fb4e8014d84950d8aeb67651b5e57713d8dcda056a65	birds_eye_view_rotation	Based on these four images (image 1, 2, 3, and 4) showing the jam bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 2, then I turn right and move forward, will I get close...	B	The question starts from the viewpoint shown in image 2, where the jam bottle is in front of a plain white wall. The target object is the white cabinet seen in another view of the same table. To answer a turn-and-move question, I first need a bird's-eye view of the starting viewpoint so I can fix the observer's current...	In the first generated bird's-eye view, the tabletop pattern around the bottle makes the starting orientation from image 2 clear. Since the question says to turn right, I apply an anticlockwise top-down rotation to the observer's facing direction. That tells me which side of the table and room will be in front after th...	The turned bird's-eye view shows the orientation after that right turn, and this new facing direction aligns with the scene where the white cabinet is ahead of the bottle. That matches the room view with the cabinet and lounge area, confirming that moving forward from the image-2 position after turning right would head...	Not supported with pagination yet
01726f7b8af34fff292afd73ba794abf1bce5255100fec3f03d99bb12c1f41d8	birds_eye_view_rotation	Based on these four images (image 1, 2, 3, and 4) showing the pink ball from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 3, what is behind me? A. Cardboard-covered sliding door ...	B	In the queried view, the camera is facing the cardboard-covered sliding door. To determine what is behind that viewpoint, I first need a bird's-eye view centered on the pink ball so the arrangement of the nearby reference objects on the bed becomes stable and the facing direction in the queried scene can be tracked wit...	The first bird's-eye view shows the pink ball with the purple cloth, patterned cloth, white iron, green device, and red-white box laid out clearly around it. Matching those bed items to the queried scene fixes the current orientation: from image 3, the camera is looking toward the side of the bed where the iron appears...	The transformed bird's-eye view is the turned-around orientation, and it matches the arrangement seen in the view with the vent hood and flexible duct above the bed. That means the direction in front of that matching original view is exactly what lies behind the observer in image 3. In that matched room view, the promi...	Not supported with pagination yet
aa609e8da9a3fda93593b9daafa471eddde831705985606816a9c5ef103aaf92	birds_eye_view_rotation	Based on these two views showing the same scene: in which direction did I move from the first view to the second view? A. Diagonally forward and left B. Diagonally forward and right C. Directly right D. Directly left	A	I need to determine how the camera moved between the two viewpoints. In the two input photos, the key anchor objects are the pink toy train and the white rectangular case on the patterned cloth. Because perspective changes can be confusing from eye-level views, I will generate a bird’s-eye (top-down) view of the first ...	In the generated bird’s-eye view of the first scene, the toy train sits near the center of the dark square cloth, and the white case is positioned toward the upper-right area relative to the train. This top-down layout gives a clear reference for where each anchor lies. Next, I will generate a bird’s-eye view of the se...	In the generated bird’s-eye view of the second scene, the toy train remains near the center of the dark cloth, but the white case is now toward the upper-left relative to the train. That shift (upper-right to upper-left around the train) indicates a counter-clockwise change in viewpoint around the setup. Using the hint...	Not supported with pagination yet
02844ad4bb56dc46468f2e194ce2613c9f119933c0548d41f5d3a56c86e01645	birds_eye_view_rotation	Based on these four images (image 1, 2, 3, and 4) showing the blue ball from different viewpoints (front, left, back, and right), with each camera aligned with room walls and partially capturing the surroundings: If I was positioned where the light-colored sofa is and facing the same direction, what would be to the rig...	B	The question asks me to imagine standing where the light-colored sofa is, while keeping the same facing direction as that sofa view. So I first compare the room cues around the blue ball across the four views: the light-colored sofa appears in one direction, the dark brown sofa in another, the white TV cabinet with a s...	In the first generated top-down view, the ball stays at the center while the patterned cloth around it is seen from directly above, which helps fix the imagined viewpoint without perspective distortion. This corresponds to standing at the light-colored sofa side and looking toward the ball. But the question is not aski...	The second generated top-down view matches the original right-side room view, the one showing the white-red cabinet beyond the ball. Using that matched original image to identify the furniture clearly, the object to the right of the blue ball from the light-colored sofa’s viewpoint is the white-red cabinet. Therefore t...	Not supported with pagination yet
37c1d957f86f45c231d474554b46cc79ae8a7fe49270a79c630ad723a1b3aab8	birds_eye_view_rotation	These four images (image 1, 2, 3, and 4) show the same scene from different viewpoints. Based on these four images: If I am standing at the same spot and facing the same direction as shown in image 3 and turn 180 degrees around, what is to my left? A. Black cabinet B. Gray chair C. Podium D. Blackboard	D	From the queried classroom view, I can see the podium on one side, the gray chair near the chalkboard, and the seating rows extending away. To answer a turn-and-direction question reliably, I first want to generate a bird's-eye view of this queried viewpoint so the current facing direction is unambiguous before applyin...	In the first generated bird's-eye view, the chalkboard runs along one side, the podium sits nearby, and the black cabinet is farther toward the corner. This top-down layout makes the original orientation from the queried view clear. The question says to turn 180 degrees around, so I now apply a 180-degree top-down rota...	The turned bird's-eye view shows the same front-of-room layout after the 180-degree change in facing. Now I apply the asked final self-relative direction: 'to my left.' Using the same ego-aligned top-down convention, a left query corresponds to a clockwise 90-degree rotation relative to the turned viewpoint. That gives...		The final aligned top-down view matches the orientation of the close chalkboard-facing original view. In that matching original view, the direction straight ahead is the chalkboard/blackboard along the wall. Since the front of that aligned view corresponds to 'left after turning around' from the queried stance, the obj...
49273e4cfad54fa93a17956080a1cc16680d43ca0e7b4084c0c81f7f3fa74583	birds_eye_view_rotation	Based on these four images (image 1, 2, 3, and 4) showing the white bottle from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: From the viewpoint presented in image 1, what is to the right of the white bottle? A. Table with cups on it B. Bed sheet with a floral ...	A	I first inspect the queried view, which is the first original image showing the bottle on the bed with the clothes rack to one side, a stool ahead, and a glass barrier on the other side. To answer what is to the bottle’s right from that exact viewpoint, I want to convert this egocentric rightward direction into a top-d...	The bird's-eye view centers the bottle on the patterned bedspread and clarifies the local layout around it. In this overhead framing, the bottle’s right side in the original camera view corresponds to a 90-degree anticlockwise reorientation when I want to compare it to another normal perspective from the room. So the n...	After rotating the bird's-eye view, the aligned orientation matches the second original image: the bottle is seen with a table directly beyond it, and several cups and containers are visible on top of that table. Because this matched original image is the normal-perspective version of the rotated layout, the object str...	Not supported with pagination yet
1c7807b359245bbb9bbfb3c2e338ccaed4f4f4073cb26f16fa558876572000bd	birds_eye_view_rotation	Based on these four images (image 1, 2, 3, and 4) showing the golden jar from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 1, then I turn left and move forward, will I get closer...	A	The question uses the viewpoint in image 1 as the starting position and asks about the white headboard, which is clearly visible in another room-facing view. From image 1, I can see the jar on the table with the surrounding room behind it, but to answer a turn-and-move question I first need the observer’s current facin...	In the first generated bird's-eye view, the shoes and golden jar are seen from directly above, which clarifies the tabletop orientation for the starting viewpoint. This top-down view lets me align the observer’s forward direction from image 1 with the arrangement of the objects on the cloth. Since the question asks for...	In the turned bird's-eye view, the tabletop has been rotated to represent the observer after turning left, so the new forward direction is now toward the side that corresponds to the headboard-facing scene. This matches the orientation seen in the original view where the white headboard appears ahead of the table setup...	Not supported with pagination yet
c2be44c7edecd5132054eca0131930c71a2efdd823784060ee2d6344fc8394f1	birds_eye_view_rotation	Based on these four images (image 1, 2, 3, and 4) showing the black leather shoe from different viewpoints, with each camera aligned with room walls and partially capturing the surroundings: If I am standing at the same spot and facing the same direction as shown in image 2, what is behind me? A. Light purple sofa B. W...	D	In the queried viewpoint, the shoe is seen from the heel end, so the camera is facing toward the back of the shoe and toward the balcony-side area with plants and railing. The other views show the same setup from additional sides, including one toward the interior and one toward the opposite side of the table. To answe...	The first generated bird's-eye view shows the shoe lengthwise, with the toe pointing upward and the heel nearer the bottom, matching the queried viewpoint where the viewer is looking toward the heel side. So the viewer in the queried viewpoint is positioned below the shoe in this top-down layout and facing upward. Sinc...	The transformed bird's-eye view represents that turn-around orientation from the same position, and this orientation matches the original view that looks toward the opposite side of the table: the one with the wardrobe area and curtains. In that matching original view, the scene ahead includes brown curtains and window...	Not supported with pagination yet
9f0e70973b7ee84eb8b4ec82b2f63570e758766da10a2079339559a88ae0299d	birds_eye_view_rotation	Based on these two views showing the same scene: in which direction did I move from the first view to the second view? A. Diagonally forward and left B. Diagonally forward and right C. Directly right D. Directly left	B	I first compare the two ground-level views and pick stable anchor objects on the tabletop: the spray can near the fold/edge of the blue cloth, the black round lid, and the green packets. The background changes a lot, so to avoid confusion from perspective distortion I need to generate a bird’s-eye (top-down) view of th...	In the generated bird’s-eye view of the first scene, the spray can sits near the central fold line between the blue cloth (left side) and the patterned surface (right side). The black round lid and a small green packet are on the blue cloth below-left of the can, while a green rectangular case and another green packet ...	In the second bird’s-eye view, the same anchors are present but their arrangement relative to the frame indicates a clockwise change: the green rectangular case that was on the right side appears shifted toward the lower-right, and the blue-cloth items (black lid/green packet) appear shifted toward the upper-left relat...	Not supported with pagination yet

Instruct-180K

Instruct-180K is the instruction-tuning data of ThinkMorph-SSL-Instruct, trained from ThinkMorph-SSL-Base. It holds 180,490 interleaved reasoning traces. Each trace has a question, one or more input images, reasoning text with intermediate images, and a final answer.

The traces are grouped into 17 tasks by what their intermediate images do. The levels follow the paper: basic visual perception, static spatial understanding, and mental world modeling (object manipulation and viewpoint synthesis).

Tasks

Level	Task	Intermediate images	Rows with images	Text-only rows
L1 Basic visual perception	`bbox`	a box drawn around the target	14,686	—
	`crop`	a zoomed-in crop of the target	7,479	—
	`jigsaw_single_image`	the pieces of one image, reassembled	19,935	—
	`jigsaw_multi_frame`	separate image pieces, reassembled into one image	20,000	—
L2 Static spatial understanding	`object_segmentation`	the objects to count, masked out	5,569	—
	`depth_map`	a depth map	6,849	—
	`marked_depth_map`	the query points marked on the image, then a depth map	2,726	—
L3 Object manipulation	`object_rotation`	the object after each camera rotation	10,811	—
	`object_editing`	the scene after the edits	1,000	—
L3 Viewpoint synthesis	`birds_eye_view`	a bird's-eye view of the scene	284	—
	`birds_eye_view_rotation`	a bird's-eye view, then re-oriented views	8,371	—
	`panoramic_completion`	the views completed into a panorama	24,247	33,631
	`object_viewpoint_render`	object masks, then the view from one object toward a point	4,639	—
	`object_viewpoint_render_oriented`	object masks, the object's 3D axes, then the view along its facing direction	1,162	—
	`object_orientation`	an object mask, then its 3D axes	8,063	—
	`marked_position_view`	the view from a marked spot after turning (rendered rooms)	6,512	—
	`exo_to_ego`	the person's own view, from an outside view	—	4,526
Total			142,333	38,157

All rows are training rows. Choose a task in the Dataset Viewer to see examples; there bbox is listed as visual_search_bbox.

Files

Path	Content
`shards/annotations.tar`	`MANIFEST.json` and every task file: all rows, no images
`shards/data-*.tar`	the 286,889 images (157 GB) in 148 tar shards
`shards/SHARDS.json`	size and SHA256 of every shard
`text_only/`	38,157 rows whose images (Matterport3D, Ego-Exo4D) are not distributed here; see Rows without images
`preview/`	the Dataset Viewer examples, one file per task

Download

hf download ThinkMorph/Instruct-180K --repo-type dataset --include "shards/*" --local-dir Instruct-180K
mkdir -p data
for f in Instruct-180K/shards/*.tar; do tar -xf "$f" -C data; done

shards/SHARDS.json gives each shard's SHA256, and MANIFEST.json each extracted file's.
To read the rows without the images, download and extract shards/annotations.tar only.

Layout and row format

After extracting the shards:

data/
  MANIFEST.json
  images/<sha256>.<ext>          shared by all tasks
  bbox/train.jsonl
  crop/train.jsonl
  ...                            one folder per task

{"id": "<sha256>", "task": "bbox", "question": "...", "images": ["../images/<sha256>.jpg"],
 "reasoning_thought_0": "...", "reasoning_image_0": "../images/<sha256>.jpg",
 "reasoning_thought_1": "...", "answer": "B"}

options appears only in tasks that keep the choices outside the question.
Image paths are relative to the row's file. id is a stable public id.

Reading the data

A trace reads in this order: the input images and the question, then reasoning_thought_0, reasoning_image_0, reasoning_thought_1, and so on, then the answer.

import json
from pathlib import Path
from PIL import Image

folder = Path("data/bbox")                      # one task folder of the extracted release
with open(folder / "train.jsonl") as f:
    row = json.loads(next(f))
inputs = [Image.open(folder / p) for p in row["images"]]
steps, i = [], 0
while f"reasoning_thought_{i}" in row:
    image = row.get(f"reasoning_image_{i}")
    steps.append((row[f"reasoning_thought_{i}"], Image.open(folder / image) if image else None))
    i += 1
print(row["question"], row.get("options"), row["answer"])

Rows without images

text_only/ holds the rows of panoramic_completion whose images come from Matterport3D, and of exo_to_ego (Ego-Exo4D). 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.jsonl names 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.
Ego-Exo4D: obtain the dataset under its license at https://ego-exo4d-data.org/. Each image record names the take and the camera. The frame-selection code will be released with the training code.

Sources and licenses

Images keep the terms of the dataset they come from; check them before you use the data. Several sources allow non-commercial research use only. Licenses below are as stated by each source.

Task	Image sources	Terms
`bbox`	SA-1B; Visual-CoT (its images come from COCO, GQA, TextVQA, Visual Genome, Flickr30k, DocVQA, Open Images, CUB and others)	SA-1B Dataset Research License; Visual-CoT: CC BY-NC 4.0, images under their original terms
`crop`, `jigsaw_single_image`	SA-1B	SA-1B Dataset Research License
`jigsaw_multi_frame`	Hypersim, ScanNet, Open X-Embodiment robot datasets, Visual7W (COCO images) and other open images	Hypersim: CC BY-SA 3.0; ScanNet: ScanNet Terms of Use; Open X-Embodiment: CC BY 4.0; COCO images: Flickr terms
`object_segmentation`	LVIS v1 on COCO 2017 images	LVIS annotations: CC BY 4.0; COCO images: Flickr terms
`depth_map`, `marked_depth_map`	RefSpatial: Open Images, CA-1M and simulated scenes	RefSpatial: Apache-2.0; Open Images: CC BY 2.0; CA-1M: CC BY-NC-ND
`object_rotation`	MVImgNet	MVImgNet Terms of Use (non-commercial research)
`object_editing`	synthetic scenes in the style of CLEVR	CLEVR: CC BY 4.0
`birds_eye_view`, `birds_eye_view_rotation`	MindCube (its images come from ARKitScenes, WildRGB-D, DL3DV-10K and self-collected scenes)	MindCube: MIT; ARKitScenes: Apple license, non-commercial
`panoramic_completion`	Realsee3D, 360+x; Matterport3D (text only)	Realsee3D: CC BY 4.0; 360+x: CC BY-NC-SA 4.0; Matterport3D: Terms of Use
`object_viewpoint_render`, `object_viewpoint_render_oriented`, `object_orientation`	COCO 2017 images	Flickr terms (COCO terms of use)
`marked_position_view`	SAT (rendered rooms)	MIT
`exo_to_ego`	Ego-Exo4D (text only)	Ego-Exo4D license agreement

The reasoning text, the intermediate images and the release layout are ours. There is no blanket license over the images.

Downloads last month: 309

Models trained or fine-tuned on ThinkMorph/Instruct-180K

ThinkMorph/ThinkMorph-SSL-Instruct

Any-to-Any • Updated about 2 hours ago • 1