yrlyrl/spatial-mmcot

Spatial MMCoT v1

Spatial reasoning data in one interleaved format: input image(s) and a question, then rounds of plan → target image, then a read-back and the answer. How much of that the text carries differs by source (supervision_kind in each source's meta, and each card says which):

• full_interleaved (cova, messytable, motif, thinkmorph_jigsaw, thinkmorph_spatial_nav, zebra_jigsaw, zebra_multihop, zebra_tetris): the upstream trace itself interleaves text and target images (drawn or rendered states; messytable's target is a held-out overhead camera photograph); the text is upstream's and was not checked against the images.
• visual_aux (ipt_mvc, ipt_pet, ipt_pt): the text comes from a separate text-only chain of thought that never saw the generated image; use these rows as image supervision, not as evidence that drawing a state helps.
• visual_only (vdrop): no reasoning text: every thought is empty.

CoVA's upstream problems have no input image, so its first drawn image is promoted to the input.

Known issues

Every source card has a Known issues section: rows with a measured per-row problem, listed in the source's reports/known_issues/ so they can be filtered out (no row was removed for them), and measured caveats that cannot be listed row by row. Read it before training on a source. The cards of cova, ipt_mvc, ipt_pet, ipt_pt, messytable, motif, thinkmorph_jigsaw, thinkmorph_spatial_nav, zebra_jigsaw, zebra_multihop, zebra_tetris report defects in what the rows teach (marked Training-signal defect), not only in how they are described.

Rows listed per issue (a row can be listed under more than one issue):

• cova: maze_label_keeps_start_end 1,140; maze_other_shortest_paths 1,076; cube_target_ahead_of_text 719; spatialeval_plan_names_answer 173; spatialeval_broken_bold_fragment 22
• ipt_mvc: options_give_answer 1,508; plan_tally_below_answer 109; plan_sees_no_object 28
• ipt_pet: plan_turn_rule_backwards 673; contradicting_answers_same_images 15; plan_repeats_donor_chain 1
• ipt_pt: plan_cites_hidden_travel_direction 3,707; plan_faces_wrong_way 2,237; plan_shared_with_other_row 817; plan_says_answer_was_given 4
• messytable: readback_excludes_same_category 28; readback_counts_beyond_target 7
• motif: count_options_give_answer 147
• thinkmorph_jigsaw: readback_opens_with_moved_comparison 172; text_states_other_arrangement 90; control_char_residue 49; target_shared_with_other_row 12; validation_target_near_copy_of_train 1
• thinkmorph_spatial_nav: plan_misplaces_start 175; validation_maze_near_twin_in_train 47; readback_claims_path_fails 23
• zebra_jigsaw: readback_names_no_option 1,077; text_says_options_not_shown 64; target_shared_with_other_row 16; question_states_answer 1
• zebra_multihop: target_repeats_earlier_target 695; readback_states_no_count 385; readback_states_other_count 13
• zebra_tetris: plan_names_answer_before_pieces 4,988; translation_clips_cells 2,242; shape_cell_count_contradicts_image 1,089; readback_names_no_option 951; target_repeats_previous_target 541; validation_target_identical_to_train 34; text_says_options_not_shown 27

Sources and licences

Each source is its own repository, with its own licence; the licence column gives each source's upstream licence, and each source card's Citation section names the upstream work to cite.

source	train	validation	target image slots	distinct target images	licence	repo
cova	3,370	127	13,066	13,021	undeclared (cova_arc rows: cc-by-nc-4.0)	yrlyrl/spatial-mmcot-cova
ipt_mvc	8,991	210	9,201	4,724	undeclared	yrlyrl/spatial-mmcot-ipt_mvc
ipt_pet	8,420	212	8,632	4,973	undeclared	yrlyrl/spatial-mmcot-ipt_pet
ipt_pt	9,110	287	9,397	5,924	apache-2.0	yrlyrl/spatial-mmcot-ipt_pt
messytable	1,490	27	1,517	1,517	undeclared	yrlyrl/spatial-mmcot-messytable
motif	20,429	636	44,558	44,525	mit	yrlyrl/spatial-mmcot-motif
thinkmorph_jigsaw	5,654	145	5,799	5,793	undeclared + ADE20K / SUN RGB-D terms	yrlyrl/spatial-mmcot-thinkmorph_jigsaw
thinkmorph_spatial_nav	5,747	170	5,917	5,917	undeclared	yrlyrl/spatial-mmcot-thinkmorph_spatial_nav
vdrop	6,748	267	7,015	1,158	cc-by-4.0	yrlyrl/spatial-mmcot-vdrop
zebra_jigsaw	10,643	319	10,962	10,954	cc-by-nc-4.0 + ImageNet terms	yrlyrl/spatial-mmcot-zebra_jigsaw
zebra_multihop	2,972	94	11,404	10,638	cc-by-nc-4.0	yrlyrl/spatial-mmcot-zebra_multihop
zebra_tetris	9,665	328	48,759	46,735	cc-by-nc-4.0	yrlyrl/spatial-mmcot-zebra_tetris
total	93,239	2,822	176,227	155,879

Distinct target images are counted by content hash within each source.

Benchmark overlap

• cova's spatialeval rows are prompts copied from the SpatialEvalLLM benchmark (yyamada/SpatialEvalLLM, ring and tree files): do not score a model trained on them on that benchmark.
• ipt_mvc: the IPT MVC AI2-THOR evaluation set draws its options the same way as this source (four consecutive counts, the correct one never the largest), so a model trained here can gain on it from the options alone; see the source card.
• ipt_pet shares no ProcTHOR house with the IPT PET-AI2THOR evaluation set (weikaih/imaginative-perception-token-pet-eval-ai2thor), but every one of that set's questions uses a template found in this source's training rows, and its renders come from the same ProcTHOR renderer: that benchmark is in-domain, not out-of-distribution, for a model trained on it.
• messytable comes from the same capture rig, objects and templates as the IPT MessyTable counting benchmark (scenes disjoint): that benchmark is in-domain, not out-of-distribution, for a model trained on it.
• motif: VisWorld-Eval ballgame uses the same task text as ball_tracking_naive, so treat it as in-domain; VisWorld-Eval multihop asks the same kind of question as manipulation_naive about CLEVR-style scenes (no shared images found by S11), so treat it as near-domain; sokoban_naive is excluded from this release (its card says why).
• thinkmorph_jigsaw: 2 puzzles whose assembled photograph is a What's Up image were removed (S11); upstream ThinkMorph Jigsaw_Assembly still contains them.
• thinkmorph_spatial_nav: mazes identical to VSP-maze evaluation items were removed (S11), but VSP-maze is in-domain (the ThinkMorph paper uses it as its in-domain benchmark for this task): it uses the same FrozenLake renderer and question text, and most of its 3x3 items have a training maze that differs only in where the start is. Only its 7x7 and 8x8 items are held out, and they are larger than any training grid (3x3 to 6x6).

Files

parquet_info_all.json merges the released sources' indexes for a mixed training run. Its keys are <source>/<split>/<file>, and the loader matches them against os.path.join(data_dir, file) exactly. Download each source into <root>/<source> (the source name, not the repository name) and either run from <root> with data_dir: <source>/train, or rewrite every key k to os.path.join(root, k) and use data_dir: os.path.join(root, '<source>', 'train'). Do not strip the <source>/ prefix: the sources' file names collide. A source whose keys do not match adds no rows to the mixed group, and there is no error.

num_used_data counts chunk files, not rows: the loader repeats a source's file list up to that many entries, so a source read with the same value as the others but twice as many files has each of its rows read half as often. Give every source the same multiple of its training chunk-file count, 128 per file as below: every file is then listed 128 times, every row group is read, and each source's share is proportional to its row groups (all row groups hold up to 128 rows, so up to each file's last, partial group, to its rows).

source	training chunk files	row groups	num_used_data
cova	1	27	128
ipt_mvc	2	72	256
ipt_pet	1	66	128
ipt_pt	2	72	256
messytable	1	12	128
motif	1	160	128
thinkmorph_jigsaw	1	45	128
thinkmorph_spatial_nav	1	45	128
vdrop	1	53	128
zebra_jigsaw	1	84	128
zebra_multihop	1	24	128
zebra_tetris	2	76	256

The rows of ipt_mvc (4 to 5 inputs), ipt_pt (3 inputs), messytable (2 to 7 inputs), vdrop (2 inputs) have more than one input image; every other source has one. A run that includes them needs a loader that reads num_input_images (the UnifiedEditIterableDataset of the IPT release, https://github.com/weikaih04/Imaginative-Perception-Token, data/interleave_datasets/edit_dataset.py, or ThinkMorph's with the two-line change shown in those cards). The stock ThinkMorph loader takes image_list[0] as the only input and trains the other inputs as generated images, without an error: on a row with two inputs and one target it trains the second input after the plan and the real target after the answer, and on a row with three or more inputs and one target it never trains the real target.

The zebra_* thoughts keep upstream's 'THOUGHT n:' labels inside <think>; the other sources have none (the zebra cards show how to strip them).

Every image is a JPEG, and no input image is larger than 512 px on its long edge (measured 2026-09-25). The conversion code (scripts/convert/) that the cards cite is not public yet.

reports/ holds the cross-source stage reports: export_summary.json, s10_summary.json, s12_summary.json, s13_summary.json, s10_label_conflicts.jsonl. The frozen S12 validation keys are in splits/ (val_keys.tsv: <source>, <corpus>::<key> for keyed policies, where <corpus> is meta.source_scene_corpus and <key> is meta.scene_id (meta.geometry_uid for zebra_jigsaw and thinkmorph_spatial_nav), with two exceptions: ipt_mvc and ipt_pet write the ProcTHOR house as scene-<N>, which meta.scene_id spells house_<N> / ProcTHOR_House_<N>, and in cova, motif, zebra_tetris and zebra_multihop rows that share an input image carry the smallest key of their group; val_members.tsv: <source>, first-image pHash for cluster policies; key_digest.tsv: per-source key digests). They are split keys, not data rows; the rows are in the per-source repositories.