phenomenon · claim-type · cross-domain-transfer · mechanism · claim-relation · replication-status · limitation — grounded in the full text| phenomenon | an ill-posed regression target: two orientation values that describe the same box but generate maximal loss against each other— “Previous methods of angle regression, including corner encoding, direct encoding and vector encoding, usually exhibit poor performance. The corner prediction approach cannot determine the direction of an object and cannot be used for pedestrian detection, for which the BEV boxes are nearly square. The vector encoding approach retains… | |
| claim-type | method— “In this paper, we present a novel approach called SECOND (Sparsely Embedded CONvolutional Detection), which addresses these challenges in 3D convolution-based detection by maximizing the use of the rich 3D information present in point cloud data. This method incorporates several improvements to the existing convolutional network architecture.… | |
| method — evidence | — “Our network usually generates ∼70k anchors within a KITTI point cloud. Unfortunately, there are usually only a few ground truths, each of which generates only 4–6 positives. This leads to an extreme imbalance between the foreground and background classes.” | |
| object of study — evidence | — “Voxel-based 3D convolutional networks have been used for some time to enhance the retention of information when processing point cloud LiDAR data. However, problems remain, including a slow inference speed and low orientation estimation performance.” | |
| research question — evidence | — “Voxel-based 3D convolutional networks have been used for some time to enhance the retention of information when processing point cloud LiDAR data. However, problems remain, including a slow inference speed and low orientation estimation performance.” | |
| contribution — evidence | — “In this paper, we present a novel approach called SECOND (Sparsely Embedded CONvolutional Detection), which addresses these challenges in 3D convolution-based detection by maximizing the use of the rich 3D information present in point cloud data. This method incorporates several improvements to the existing convolutional network architecture.… | |
| transfer | from 2D object detection: the focal loss, carried from single-stage 2D image detection into 3D point-cloud detection to handle the same foreground-background imbalance— “Our network usually generates ∼70k anchors within a KITTI point cloud. Unfortunately, there are usually only a few ground truths, each of which generates only 4–6 positives. This leads to an extreme imbalance between the foreground and background classes.” | transfers[0] |
| mechanism | sampling ground truths from a database objects and their enclosed points are collected once into a database, then pasted into other point clouds during training, which multiplies the labelled objects per scene and simulates each object in new surroundings— “The major problem we encountered during training was the existence of too few ground truths, which significantly limited the convergence speed and final performance of the network. To solve this problem, we introduced a data augmentation approach. First, we generated a database containing the labels of all ground truths and their associated… | mechanisms[0] |
| mechanism | collision rejection after sampling pasted objects are removed when they intersect existing ones, so the augmentation cannot produce a scene that could not physically exist— “To avoid physically impossible outcomes, we performed a collision test after sampling the ground truths and removed any sampled objects that collided with other objects.” | mechanisms[1] |
| mechanism | per-object rather than per-scene noise each ground truth and its points are transformed independently instead of the whole cloud being transformed together, so the augmentation varies objects relative to their scene— “To consider noise, we followed the same approach used in VoxelNet, in which each ground truth and its point cloud are independently and randomly transformed, instead of transforming all point clouds with the same parameters.” | mechanisms[2] |
| extends | VoxelNet, which combines raw-point and voxel feature extraction end to end— “Recently, a new approach called VoxelNet has been developed. This approach combines raw point cloud feature extraction and voxel-based feature extraction in a single-stage end-to-end network. It first groups point cloud data into voxels and then applies linear networks voxel by voxel before converting the voxels into dense 3D tensors to be… | claim_relations[0] |
| critiques | prior angle encodings - corner, direct and vector— “Previous methods of angle regression, including corner encoding, direct encoding and vector encoding, usually exhibit poor performance. The corner prediction approach cannot determine the direction of an object and cannot be used for pedestrian detection, for which the BEV boxes are nearly square. The vector encoding approach retains… | claim_relations[1] |
| builds-on | the focal loss introduced for single-stage detection— “Our network usually generates ∼70k anchors within a KITTI point cloud. Unfortunately, there are usually only a few ground truths, each of which generates only 4–6 positives. This leads to an extreme imbalance between the foreground and background classes.” | claim_relations[2] |
| sense | 'orientation error' was measuring the ANGLE and not the BOX - two angles describing one box produced maximal loss— “Previous methods of angle regression, including corner encoding, direct encoding and vector encoding, usually exhibit poor performance. The corner prediction approach cannot determine the direction of an… | senses[0] |
| sense | the new loss treats boxes with opposite directions as the SAME by design, and a separate classifier recovers the direction Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | senses[1] |
| sense | the direction target is defined by a sign test on yaw: positive above zero, negative otherwise Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | senses[2] |
| sense | 'state of the art' here is dataset-relative: 90% AP for 2D car detection against 15% for 3D image-based Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | senses[3] |
| sense | 'real-time' is stated as a run time, not a claim - 20 fps for the large model and 40 for the small Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | senses[4] |
| control | the same detector with dense convolution instead of sparse, to attribute the speed gain — held Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. — rules out a speed gain produced by the rest of the detector rather than by sparse convolution Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | controls[0] |
| control | the angle encoding used by AVOD and ComplexYOLO, run as the comparison for the new angle loss — held Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. — rules out a new angle loss credited with an improvement that the encoding used by the baselines would also have produced Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | controls[1] |
| control | a physical-plausibility check on the augmented scenes — held Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. — rules out a detector trained on augmented scenes containing objects that could not physically coexist, which would teach it geometry no sensor will ever show it Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | controls[2] |
| control | a validation-set assessment of the larger model — absent-declared-unreported Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. — rules out over-fitting of the larger model to the test set - the validation-set assessment that would show it exists only for the smaller model, because access to the test server was limited Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | controls[3] |
| note | VoxelNet, which combines raw-point and voxel feature extraction end to end — keeps the voxel approach and replaces the dense 3D convolution that made it too slow Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | claim_relations[0] |
| weight | VoxelNet, which combines raw-point and voxel feature extraction end to end — 0.95 Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | claim_relations[0] |
| note | prior angle encodings - corner, direct and vector — each is rejected for a different failure, and the replacement is a loss that changes what is being measured rather than how well Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | claim_relations[1] |
| weight | prior angle encodings - corner, direct and vector — 0.95 Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | claim_relations[1] |
| weight | the focal loss introduced for single-stage detection — 0.95 Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | claim_relations[2] |
| audit | anchors —contrasted-with→ ground truths — Audit 2026-09-28: DOUBTFUL - read against its full span and kept, but the span contrasts the NUMBERS of anchors and ground truths; contrasted-with is loose. Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concept_edges[2] |
| differs in | anchors —contrasted-with→ ground truths — count - about 70,000 anchors against a few ground truths each yielding 4 to 6 positives Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concept_edges[2] |
| weight | anchors —contrasted-with→ ground truths — 0.95 Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concept_edges[2] |
| outcome | control — held Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | controls[0] |
| outcome | control — held Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | controls[1] |
| outcome | control — held Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | controls[2] |
| outcome | control — absent-declared-unreported Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | controls[3] |
| limitation | performance is lower for pedestrians and cyclists and for bird's-eye-view detection, which the paper states in its own conclusion Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | limitations[0] |
| limitation | pedestrians and cyclists give fewer points per instance and higher instance density, so they are more easily confused with clutter Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | limitations[1] |
| limitation | test-server access was limited, so the larger model was evaluated on the test set only and the validation assessment covers the smaller model alone Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | limitations[2] |
finding · quantity · variable · definition · key-term — grounded in the full text| finding | sparse convolution gives a large speed gain over dense convolution at slight cost in performance Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.findings[0] |
| finding | the new angle loss outperforms the vector encoding used by two prior detectors Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.findings[1] |
| finding | the method reaches state-of-the-art results using LiDAR alone Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.findings[2] |
| finding | performance remains lower for pedestrians and cyclists than for cars Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.findings[3] |
| quantity | 90% AP for 2D car detection against 15% for 3D image-based car detection — Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.quantities[0] |
| quantity | a factor-of-4 training speedup and factor-of-3 inference speedup against dense convolution — Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.quantities[1] |
| quantity | about 0.025 s per frame for the small model on a GTX 1080 Ti — Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.quantities[2] |
| quantity | 20 fps for the larger model and 40 fps for the smaller — Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.quantities[3] |
| quantity | about 70,000 anchors per point cloud against a few ground truths giving 4 to 6 positives each — Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.quantities[4] |
| variable | orientation Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.variables[0] |
| variable | anchors Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.variables[1] |
| definition | angle regression Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.definitions[0] |
| definition | direction classifier Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.definitions[1] |
| concept | sparse convolution — not defined in this paper's own words; convolution over only the occupied voxels, so cost tracks occupancy rather than volume Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concepts[0] |
| concept | angle regression — not defined in this paper's own words; predicting a box's orientation as a number - the thing this paper argues was measured wrongly Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concepts[1] |
| concept | corner encoding — not defined in this paper's own words; an orientation encoding that cannot determine direction, and fails on near-square pedestrian boxes Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concepts[2] |
| concept | direction classifier — not defined in this paper's own words; the binary output added to recover what the new angle loss deliberately discards Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concepts[3] |
| concept | ground truths — not defined in this paper's own words; the labelled objects; too few per point cloud, which is what the augmentation addresses Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concepts[4] |
| concept | anchors — not defined in this paper's own words; about 70,000 per point cloud against a handful of positives - the source of the class imbalance Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concepts[5] |
| concept | VoxelNet — not defined in this paper's own words; the state-of-the-art predecessor, accurate and too slow for real time Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concepts[6] |
| concept | KITTI — not defined in this paper's own words; the benchmark every claim in the paper is measured on Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | concepts[7] |
| average precision (AP) | Quotation withheld — this paper is not ours to reproduce, and the page has already quoted as much of it as the rights layer allows. The tag, its field and its confidence stand; the sentence is at the publisher. | content_tags.statistics[0] |
When we have changed our mind. Some entries in the table begin with a word in capitals and the reason for it. EVIDENCE CORRECTED — the finding still stands, but we had quoted the wrong sentence from the paper, and the quotation has been replaced with the right one. RETRACTED — the entry itself was wrong; it stays on the page, marked, with the reason, rather than being deleted. WITHDRAWN AND REPLACED — the entry was wrong, and a corrected version of it appears in the entries below. Each mark carries its own reason, in brackets, because the reasons differ from one entry to the next. An entry we have WITHDRAWN is removed from this page and kept in the record, with its reason. A line reading Audit and a date is a later re-reading of the entry against its whole sentence: DOUBTFUL means it was kept, and the reason says what is loose about it. Quotation marks inside the tagger's sentence. Where a tag is the tagger's own sentence, a phrase in quotation marks inside it is borrowed from the paper — the tagger's wording around it is not. The paper's own sentence is the evidence quote for that row; on this page it is withheld with the article, and the article is at its publisher. Nothing inside quotation marks anywhere on this page is ours: if it is quoted, it is the paper's. Whose words a concept carries. A concept entry records a term the paper itself uses, so the term, the sense given for it, and its evidence are the paper's own words, quoted where the paper fixes them. This is the opposite of a sense entry in the tag table, which is the tagger's sentence about how the paper uses a term. Where a concept entry could not be grounded in the paper's words, the entry says so on its own row.
Tag layer © the Meta-Tagging Project, CC BY 4.0. The paper text is © its authors/publisher
and is linked, not reproduced. Excerpts shown are short quotations for identification and verification.
· back to index
· concepts · authors
· עברית
© 2026 Shir Sivroni · ORCID 0009-0000-2597-2824