hahaws777/six-view-active-perception

Rendered RGB images for closed-set active object recognition: one movable camera chooses which named view of an object to look at next. Every scene is a group of six 512×512 views of one object, rendered in NVIDIA Isaac Sim 6.1.0, under normal light, low light or partial occlusion.

• 38,160 images, 6,360 scenes, 156 independent 3D models (Objaverse-LVIS).
• Six classes: mug, bottle, can, bowl, camera, laptop (label index in this order).
• Splits are object roles (methodology v3 of the project): no model appears in two splits.

Split	Models	Normal light	Low light	Partial occlusion	Images	Role
train	72	8,640	8,640	8,640	25,920	recognizer training; policy-training pool
policy_val	18	1,080	1,080	1,080	3,240	BC checkpoint selection; update selection within each PPO variant
recognizer_val	12	720	0	0	720	recognizer early stopping / checkpoint selection
calibration	12	720	0	0	720	temperature scaling only
dev_test	12	720	720	720	2,160	evaluation (repeated development benchmark)
final_test	30	1,800	1,800	1,800	5,400	evaluation only (see note)

In the project's methodology-v3 protocol, training and every selection use Normal light only; the low-light and occlusion renders of train and policy_val are kept for other studies, and the shifted conditions are evaluation conditions for dev_test and final_test. recognizer_val and calibration (24 models added in 2026-10) exist in Normal light only. Note on final_test: these 30 models were held out until 2026-10-19 and have since been evaluated several times by the project; they are not an untouched test set any more.

Scenes and views

Each scene fixes one object pose and a start bearing and distance. The six views are named absolute camera poses: FRONT (the start), LEFT, RIGHT, BACK at the same radius, TOP above, and CLOSER on the FRONT bearing at half the radius. Group the six views of a scene by scene_id (and condition). Shifted scenes are paired: a low-light or occluded scene has exactly the same object, start, camera poses and intrinsics as the normal scene with the same scene_id; only the lights (scaled to 10%, auto-exposure off) or a fixed blocker change. Models have 20 starts in train and 10 in every other split.

Fields

image, label, class_name, role, instance_id, source_uid (Objaverse UID), family_id (near-duplicate family; never shared across roles), release (which collection rendered it), condition (iid_normal, low_light, partial_occlusion), view_id, scene_id, start_index, seed, camera_position_m, camera_orientation_wxyz (world frame, Z up), camera_radius_m, start_yaw_degrees, light_scale, occlusion_enabled, image_sha256.

from datasets import load_dataset
ds = load_dataset("hahaws777/six-view-active-perception")                        # all roles
train = load_dataset("hahaws777/six-view-active-perception", split="train").filter(lambda r: r["condition"] == "iid_normal")

Storage: 7 Parquet shards under data/ (images embedded), so new collections add a few files rather than thousands. The earlier per-PNG layout (train/validation/test folders, v2 + v4 renders only) is in this repository's history.

How it was made

Models come from Objaverse-LVIS (CC BY / CC0 only), were screened for duplicate geometry and author families, converted to USD, normalized to a 0.25 m largest dimension and visually accepted before any split was frozen. Images were rendered headless with RTX rendering, no anti-aliasing and fixed exposure; every image's SHA-256 is recorded. Collection, quality reports and the experiments are in the project repository (chenytya/active-perception-inspection).

Intended use and limitations

Research on view selection, multi-view recognition and robustness to lighting and occlusion. Synthetic renders on a plain background; one severity per shift; 2–5 test models per class, so per-condition estimates move by several points per model.

License and attribution

The images are renders of third-party 3D models and are released under CC BY 4.0. Each source model keeps its own license (153 × CC Attribution, 1 × CC0 Public Domain, 2 × by). When you use this dataset, credit the model authors below and Objaverse:

Deitke et al., Objaverse: A Universe of Annotated 3D Objects, CVPR 2023. Objaverse is released under ODC-By v1.0.

Class	Role	Model	Author	License	Source
mug	train	Coffee Cup	linmardon	CC Attribution	45381c7f838540d88ed85cd2f00cb35b
mug	train	Mug	Hac_Pc	CC Attribution	4fbbdf9506be4b6ea0d4ce74007a37f8
mug	train	Mug	huber68	CC Attribution	52d54d64e2544ace8a41d33d2ee03fdc
mug	train	Wood mug	KulerRuler	CC Attribution	91c5283b27c74583900d5e26e2fcd086
mug	train	Mug Handle	klen.copic	CC Attribution	9321c45cb9cf459f9f803507d3a11fb3
mug	train	Olut tuoppi ilman olutta 2	rojuvil	CC Attribution	95fe6a8daebb4ea5846ace8c0806d108
mug	train	TAZA HT NO 3	TAZA	CC Attribution	a17de88a9464410fa2c8f550ab04749e
mug	train	Pink Mug	benSkope	CC Attribution	c25119c5ac6e4654be3b75d78e34a912
mug	train	Grey mug	BranislavJanosik	CC Attribution	c666d90ad57e44f1baf90e5ec062a152
mug	train	Cup	gmodfan2005	CC Attribution	e408136e84414784a142e40eb876bbed
mug	train	Komi_cup	GabeLdc	CC Attribution	e5e87ddbf3f6470384bef58431351e2a
mug	train	Rusted Over Mug	Nick	CC Attribution	fa7abc1ab50c484db3323ccbf7f8514b
mug	policy_val	Singapore Coffee Cup Polycam AI Photos	sressler	CC Attribution	1b36172e3e254b46a973f0132c4b2e08
mug	policy_val	Personal Mug With Logo/Crest	spencertodd	CC Attribution	922cd7d18c6748d49fe651ded8a04cf4
mug	policy_val	Mug	Artem Goyko	CC Attribution	ad2546804e8845b09e5db7c6db16b704
mug	recognizer_val	Wooden Mug	CHRISRAM	by	047462f5fdc244339511f87f4df49446
mug	recognizer_val	Mug with Snake Decal	gottalovehana	by	82e943e3cfac4da5a20d189ddbd6abdf
mug	calibration	"Włocławek" mug from the house of Szymon Kluger	Virtual Museums of Małopolska	CC Attribution	543c9a06fc8f47648f440b2c36b354d0
mug	calibration	Scene	wmich50	CC Attribution	a564a11174be455bbce4698f7da92fdf
mug	dev_test	LeCreuset - Mug	mwheatley	CC Attribution	6702d5d92c9c416eac5cc99ee383ff46
mug	dev_test	Nordic Mug	biliam	CC Attribution	6734e6f2fc6e4f80a887e6cfc9c92cc4
mug	final_test	Mini Mug	Gustavo A. Bertran	CC Attribution	2b3e80b25a434b15b31cd5078b7f059a
mug	final_test	Cup	gorbovski	CC Attribution	31b7048a2fbe4a038a9f216f1a966bb2
mug	final_test	Coffee Mug	alexababin	CC Attribution	3a87a55210db4fd9a27565671824f894
mug	final_test	Chocolate Cup	cleisonrodrigues	CC Attribution	44b45ae8126a4358ac49321c538e2c31
mug	final_test	Jack Skellington cup	AoBlue_Moon	CC Attribution	6142a5564edb4a7ba1c948673460afe0
bottle	train	FH Erfurt Brause	J0ye	CC Attribution	0d8c95eb3acc4da29543dd8c8ced2688
bottle	train	Garrafa De Vinho	KaremCaram	CC Attribution	5369d70668b441eda59a5e1dba46ef1c
bottle	train	Coca Cola Bottle	Tyler Hales	CC Attribution	568e73d81cb94e4bbfe8c3e0cc334bab
bottle	train	Garrafa De Vinho	flaviaale1818	CC Attribution	7b2f7373c6904a32a76a9b9730983ccf
bottle	train	B-4 (500ml)	Athif	CC Attribution	9337ea6cf4e942359f290c69c77e132d
bottle	train	Wine Bottle	ReeceLaudat	CC Attribution	b60b4974c3724f478cfe7899a4896120
bottle	train	Tequila Bottle	louis.rapine	CC Attribution	c1b28950a23846aeb02a3e2cfbf590eb
bottle	train	Wine Bottle Burgundy Style	n30hrtgdv	CC Attribution	d03766ba61f2468aa95709ce84b5a4e4
bottle	train	Bottle Ryan Export	RyanBurt	CC Attribution	e542ab7c0c8a4a38a1f8d03df94f0fe0
bottle	train	Waterbottle	dyran	CC Attribution	ef4d26f737fd4ce88e39387e419de1b3
bottle	train	Bottle	A109082010	CC Attribution	ef94916c83e441d483e6e3b9647d274b
bottle	train	Golden Goat Beer Bottle	FrodoUndead	CC Attribution	f55671ec28d74e28b5f28e9fd906d8c8
bottle	policy_val	Nuka Cola Bottle	thecakeface	CC Attribution	1c2a9a7e360b4d2892654f3f8998ef85
bottle	policy_val	Low Polly Camping Canteen	Gabriel Tarzia	CC Attribution	2f6f4c5799c84ce1b05f6de2e04d5a21
bottle	policy_val	Wine Bottle	Even Haaland	CC Attribution	fb8ba6812b6847f8b7f26dc319c07f93
bottle	recognizer_val	GARRAFA	Kaio.Da.Fonseca.Pacheco	CC Attribution	13f29fef23f5432f8e179118ae3c14c7
bottle	recognizer_val	Bagnoschiuma Felce Azzurra del 1979	Archeoplastica: il museo dei rifiuti spiaggiati	CC Attribution	40f7bac9be754b928f24c52d9f6b340e
bottle	calibration	Beer Bottle	Louis	CC Attribution	7f2daeea9eea45f8a9ef854661e99ee4
bottle	calibration	Wine	Daniele Caccavale	CC Attribution	d81bcea0168440f486e67b0ba9caf95a
bottle	dev_test	Ipa15	AL Studios	CC Attribution	328a3c9a6be7472eb217d755466c42c9
bottle	dev_test	Beer bottle	r3mnantt3D	CC Attribution	4049954e376d4fae9e9fac0b6397ab26
bottle	final_test	Айс Tea	Qoodrat	CC Attribution	13ef8c3671204fd3924b0db9a05b21aa
bottle	final_test	Bottiglia-video2	salahzar	CC Attribution	87d061ccab964209953b1622dee66182
bottle	final_test	Vodka (Tuica)	emypustiu	CC Attribution	a3a17b679da64cbf9f30e68336bc7af5
bottle	final_test	Bottle01 0.15x0.15x0.6	moyicat	CC Attribution	a76ab0a3f26745a1812694187f4c059c
bottle	final_test	Low-poly Wine Bottle	nimitkansagra	CC Attribution	edb077bdea8446b89b9af51cbae345e0
can	train	fiesta corn	House Doctor	CC Attribution	04c9a4a2ea5442e68aa4889e92b63d14
can	train	Can of soda	Gustavo Martins	CC Attribution	117490ef21d249fba0f69bdb223523f9
can	train	Pabst Can Reduced Baked	Igor.Bondarenko	CC Attribution	204f6d75b01f48a7810e00f58cdbfc53
can	train	Gambrinus Bier	DarkyDark_	CC Attribution	33a68c221d3f460cb1cb1f78b9fe0bb7
can	train	Blikje	scythar	CC Attribution	350f3337a3e6484291bdf29537b54689
can	train	Tinned-Tomato soup	WENGA	CC Attribution	4489a6b448a74f6e93e3a721bcb9eb12
can	train	Drink Can	SamMonahan	CC Attribution	4b6af44cf17d41eeab1076c3b99e4f5d
can	train	Beercan	Versaali	CC Attribution	5235126da6d74a5a81675cd2abf5d6a2
can	train	Sunder Soft Drink	BitBandit98	CC Attribution	b959e72cb7db41fb9ca7428e635d95ff
can	train	Soda Can	Nicosama	CC Attribution	bf7902fb64644da3beb0ce1fce07f4aa
can	train	3D Sketchbook 2 - Can	cholec	CC Attribution	c49fe9a654a444648230d73b7ded05e6
can	train	Poly can	jasminemayrie	CC Attribution	c532c913264741a1a6908d13a600895d
can	policy_val	Peapsi Can (Veggie Tales)	liamandnico	CC Attribution	2ecc060e9a3c4615971a07aa4ead359a
can	policy_val	Cola Can	klaxoneer	CC Attribution	32c47e171fc54617aef4c5696f27a7a5
can	policy_val	Spray Paint Can	wendyanna	CC Attribution	4c8af55af44c4e888fa9e3ca037f65e3
can	recognizer_val	Spray Can - Textures	ipattycake	CC Attribution	10ac5e64861a4cb4868a244d08496b97
can	recognizer_val	A Realistic Textured Can	Jingfu Chen	CC Attribution	476b23ffca3f43b2a8cad354a35072eb
can	calibration	Canned Fecies (Poo Tin)	unfa	CC Attribution	663b5edc92a543668c1b602981e724a4
can	calibration	Lubricante 6 en 1	badwolfmx3	CC Attribution	e867291098e44a55920a1f59fc69dfd4
can	dev_test	Blikje_01	RubenHennessy	CC Attribution	1df6f3e239ad46d191b6d51e04fde3b8
can	dev_test	Soft-drink Can	maypassamon	CC Attribution	b521c9a1162d45d2b4f138206e114fd2
can	final_test	Dharma tin can	Simon Laisné	CC Attribution	0e1ae51d32784728bb61d3f24e78573e
can	final_test	Latinha	Henrique Godoy	CC Attribution	1362fdf6c7d341a48efc217b71b69e1e
can	final_test	Voltage	Thomas.Cole	CC Attribution	34822be490d847059f91481ab91bd2a2
can	final_test	Soda can from Poly by Google	IronEqual	CC Attribution	4e5d012176e5459d9b8cad929999072e
can	final_test	EinScanPro2XP CAN TEST	BIANKA ENGINEERING PRIVATE CHANNEL	CC Attribution	b95b68673c824c8ebb15e16ffe5e1449
bowl	train	Cuenco2	CyCulture	CC Attribution	0aaf8989743c4cf587710648ef6ed05f
bowl	train	Scodella Legno	Meshes3D	CC Attribution	111fc5314dc64506b0a058fd48d67637
bowl	train	Bowl test. Hi - spec	Noz'aj	CC Attribution	11a459d53cd04ff69da21d96ce848aec
bowl	train	木碗2 862.316	Wushen Banner Museum	CC Attribution	1e23d9b649a643528ce00de750e858de
bowl	train	Fruit Bowl	tomaszura	CC Attribution	613fc896899349e4911ccce2f94e63ee
bowl	train	Mi_102_2	Archaeological 3D virtual museum	CC Attribution	736b53104d1d4d1b8c89f8f1b599dc07
bowl	train	2732331	Archaeological 3D virtual museum	CC Attribution	7f8004fe5bf2480e836011ba70d2c423
bowl	train	2755050	Archaeological 3D virtual museum	CC Attribution	821f57d137454002bc0211d5ee58b6b1
bowl	train	Peruvian Bowl	Beth Fischer, Williams College Museum of Art	CC Attribution	826b0539e9394cc3ada0c2cd7faae5f4
bowl	train	Cuenco, cerámica granadina - Fajalauza-HD (29)	Grupo PRINMA UGR	CC Attribution	a6995104b99c4399a20bc4e0031fbf8b
bowl	train	Momo-gata Japanese Tea Cup	Melina.Thadea	CC Attribution	bad55ce1183a4965896446aa526c7ac2
bowl	train	Decretive Medieval Bowl (Free)	wolfgar74	CC Attribution	f1eccbd01c884fe28ec65ce699cf7b81
bowl	policy_val	Cherokee Complicated Stamped Pan (2176p14)	RLA Archaeology	CC Attribution	68655c9cf95642efa50c98f895c7a449
bowl	policy_val	TazonCilindrico	EmpDevTeam	CC Attribution	ede795f1c6f74a99b6952183f340fcb8
bowl	policy_val	Olive Wood Bowl Comparative Test (Metascan)	Digital Archaeology Laboratory	CC Attribution	efe008d8da0a4d58a158a191d713cbc4
bowl	recognizer_val	4-cleaned	Kristina.Ilieva	CC Attribution	15b103f822ef4d92a386e95746a9aebe
bowl	recognizer_val	Vasija Precolombina MEMAB	ProyectoSCAN	CC Attribution	610e117860ed43d4bb0dbca1099eecdb
bowl	calibration	Clay bowl 16th century	Viktoryia_Makouskaya	CC Attribution	29a9f10274f04aedb1d5103a08ad011a
bowl	calibration	Vas Mic M	Institutul Național al Patrimoniului	CC Attribution	9e297d68214143b4a36c7514ee45ef7a
bowl	dev_test	Stylised Bowl	deangothard	CC Attribution	4f6e6b0d481a48fab1cecd0700d9846c
bowl	dev_test	Catawba Earthenware Bowl (2554p3022)	RLA Archaeology	CC Attribution	da0ddb490f214ed9a5c4afddad1aea27
bowl	final_test	Cuenco restaurado	Álvaro Rodríguez	CC Attribution	0c5f7031737b4f14a3855930f02226a4
bowl	final_test	Bowl (2020)	Agecroft	CC Attribution	40c81b7afcc344648c599488c587e4a4
bowl	final_test	Bronze bowl	Världskulturmuseerna	CC Attribution	8ef413e5235e42e4982f81c0d07448a4
bowl	final_test	縄文中期初頭椀形土器（東金市鉢ヶ谷遺跡）　観察記録3Dモデル	arakiminoru	CC Attribution	a926d5a38d4842bda586c40ad3d482ac
bowl	final_test	Cuenco Waru	Ministerio de Cultura Perú	CC Attribution	e5c6cae2c71b45fe8fee9657af246c2c
camera	train	Vintage Camera	KSFX.NO	CC Attribution	26af80179971467ca67a3edd8ef91343
camera	train	Camera - Digital Point 'n' Shoot	MozillaHubs	CC Attribution	2a09a3dc75364c8c84b25d2cc235cb9b
camera	train	Boling Andrew Project3 ART2-O 1702	xKage21x	CC Attribution	34657bbeb0ec40ca89cea6775c550e0f
camera	train	Leica IIC Rangefinder Camera	frankebones	CC Attribution	44d5052c398248ddb3e4d521cae035a9
camera	train	Boling Andrew Project2 ART2-O 1301	xKage21x	CC Attribution	48ea397d2cd64bc4a39f4aa98acd29a7
camera	train	Kodachrome Camera	10779505	CC Attribution	4b3a4cffbbe4499b9e69eaf3157abf06
camera	train	Retro Polaroid Camera	McManus Media	CC Attribution	89e655a7ef994d5c8360766d59b5921f
camera	train	Stary aparat płytowy IBSO	Fundacja Collect	CC Attribution	a0671f81b7574c619a8d22d8dc5ff7a1
camera	train	camara	Frenesí.multimedial	CC Attribution	cb973ddaa1db485c824437d10aea2fb5
camera	train	2nd Draft with details	Feetra	CC Attribution	da8efc70f6f4411dbaca3928136d7bcc
camera	train	PhotoCamera	Unknown_Space_Marine	CC Attribution	f436b41f927941328b9cac832afae3e9
camera	train	Argus Camera	Virtual Museums of Małopolska	CC0 Public Domain	f9112ea4c15043ebbb24fa121ffef920
camera	policy_val	Leica M2 Summicron 35	RetroVertex	CC Attribution	6869e7b449cf4196a03a22ef84d87bed
camera	policy_val	Vintage Camera	niallplatt	CC Attribution	9aa8d6f4d32a4d61876408a27ea4acea
camera	policy_val	Camera Modelling	Damla Yagli	CC Attribution	f49a08efb8a544efb98d4d806b0d1c27
camera	recognizer_val	Camera_Leica_low_poly	SIgn_man	CC Attribution	072a06f2900a4807865dc9aa90f28ad2
camera	recognizer_val	курсовая работа "Фотоапапрат".	Plavich	CC Attribution	384ae2f1edf44f76849cfbc0321aace4
camera	calibration	Vintage camera - Certo SONNY - Dolly Vest Pocket	DonCoasn	CC Attribution	0e0b88c0baf8434fbd236ee93dc8d107
camera	calibration	Vintage Camera	Klasy	CC Attribution	c9f7688b9b0d4e2ea8e660c2b08de980
camera	dev_test	Canon AE-1 Program 35mm Film Camera	Marc Sawyer	CC Attribution	03b0ac7d99c44197a09640179f360f3c
camera	dev_test	Yashica Mat - 124 G Camera	Dungar Singh Balecha	CC Attribution	31564ba77f5a468c911c1573869b5f69
camera	final_test	Leica	Alex Voev	CC Attribution	46de6f5c28cf45d1bcfc2f634f921b5f
camera	final_test	Camera Zorki - 4	Dmitry Siromaha	CC Attribution	62829a25b4634d63a2ba2b0220335b3e
camera	final_test	Smith Michael Project 483-L 17014	Michael Smith	CC Attribution	81b3eaf9958b444781323c733255db0e
camera	final_test	Kodak camera	Fujihiko Matsumoto	CC Attribution	84a8514571034064862b36e61a026b61
camera	final_test	Video camera	Kesenkai	CC Attribution	8d89bff741d1486398a3a1bab4626abe
laptop	train	Laptop	sampurna99	CC Attribution	4db0ba558ebf4978ae925ca4b1cf9477
laptop	train	lenovo	elyas_baloch	CC Attribution	565aea3028e54b4883e89577c971a43c
laptop	train	2_Laptop	SergeyB	CC Attribution	6c832989ee684551a3a3508ee90cdda9
laptop	train	Laptop	asteens	CC Attribution	79d7d10b30b14ae09f2ef89872868560
laptop	train	Laptop Dell Latitude 5400	sabrito84	CC Attribution	7af387223fe14d6ba4cd725bc637cc8a
laptop	train	Laptop	sebastianquijas	CC Attribution	8a5b92a5cd9b48f9b3c471b81567423c
laptop	train	Laptop	CG Soul Motion	CC Attribution	a14d471ffda04d38a1910b9ef87e8dff
laptop	train	Mini Gaming Laptop	3D Y.A.	CC Attribution	aa0be973535e48c4b3b15d50f1df176a
laptop	train	Laptop(JPkey)	kerohiko	CC Attribution	b1dcc2e6d6f54aaa90c15d9092d0a32e
laptop	train	#3DST24 (Not that) old mac	Victor	CC Attribution	e6225e5e006842dcb9214bef8041850b
laptop	train	gaming_laptop	sabrito84	CC Attribution	e92205e389f24d75811f463405a33bfe
laptop	train	PC1	laldskin	CC Attribution	f84beda470b540e998e8a1dcfec93336
laptop	policy_val	EnergyOS notebook	cheb_max	CC Attribution	a5483e74a73143cdb3a453ebd167ce06
laptop	policy_val	laptop	malikalberto	CC Attribution	c76723abab6e4e2696bf7e3e1a27e861
laptop	policy_val	Mac Book 1	Little Max bnz	CC Attribution	f2f85fc7c4df416e86fdb701901466d1
laptop	recognizer_val	Laptop	MelvinMiguel	CC Attribution	6f4a0b64512f412aa46403796201ac1c
laptop	recognizer_val	Baked 3d Shortcut	funkypixie	CC Attribution	933b943b67664ab692052bdceaf11f4e
laptop	calibration	Surface Pro Black [Download]	Wacky	CC Attribution	4c23ef1cb8d2476cae143f3025be9857
laptop	calibration	Minimal Laptop	borovikovworks	CC Attribution	e1211c160a9a414a9c4788a1bd4b0c2c
laptop	dev_test	PC DELL G5 5590	CONTxE	CC Attribution	2cda50e73f69438598eabf8259e27cb7
laptop	dev_test	Laptop	Thor.Verstraeten	CC Attribution	4dea8c988c6141acad8c2217ad739a02
laptop	final_test	NoteBook no estilo de ps1	Moustache_Cat	CC Attribution	2df262e6815f4c2eb7572607ae5e0321
laptop	final_test	CyberPunk Laptop	Vivien Deroche	CC Attribution	8330c89e8431451db3aa9d15c5836ddf
laptop	final_test	LAPTOP 2	rokaye	CC Attribution	9e37ff6c22094e91a81e9e3be1e32b80
laptop	final_test	Nebula Learning Device	Nebula	CC Attribution	b87137a7189a46679de99a0b09fc9e7e
laptop	final_test	laptop	Fedor KoNsTaNtInOv	CC Attribution	d62cdacba902477ea16ad2268627fdf4