- Rating
- 3
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes KIO-World, a sparse occupancy forecasting model conditioned on historical multi-view images and future ego trajectories. It introduces a shared kinematic state for each query anchor, which supports motion-compensated historical feature sampling and soft regularization of future anchor centers. It also proposes Hierarchical Residual Point Refinement (HRPR), combining parent-point inheritance, anchor-level transformations, and point-wise residual updates. The paper evaluates two model sizes on Occ3D-nuScenes and reports improvements over corresponding SparseWorld-TC baselines. Additional experiments cover RayIoU, per-class performance, kinematic-state sharing, HRPR transfer, and longer prediction horizons.
Critical strengths
S1. The main architectural choices have direct experimental support. I find Table 10 particularly valuable because it tests the proposed sharing scheme against direction-specific and horizon-specific states. The shared variant achieves 28.2 mIoU and 51.8 IoU, compared with 27.8/50.7 and 27.2/49.9, respectively. This supports the usefulness of sharing a common motion reference across temporal directions and prediction horizons. Table 4 provides complementary evidence for HRPR through improvements on both OPUS-T and SparseWorld-TC. Together, these experiments establish more than an improvement from combining several unexamined components.
S2. The reported improvements are consistent across model sizes and complementary metrics. Both model sizes improve over their corresponding ResNet-50 SparseWorld-TC baselines. For the Large model, average mIoU increases from 26.4 to 29.4 and IoU from 49.2 to 52.3. The RayIoU evaluation strengthens this result: for the Small model, average RayIoU increases from 28.5 to 31.1. The per-class analysis also reports improvements on small movable categories. I consider this combination of evidence meaningful because it supports the method across multiple evaluation views and reduces the concern that the voxel-IoU gains arise primarily from overly thick occupancy predictions.
Critical weaknesses
W1. The attribution of the future-regularization gain remains incomplete. Table 11 shows that adding the future-center regularizer, L_reg, alone changes mIoU/IoU from 27.9/50.4 to 27.9/50.6, while combining it with the static-scene prior, L_static, yields 28.2/51.8. The authors appropriately acknowledge the importance of the static-scene prior. However, without a variant that enables L_static and disables L_reg, the additional benefit of the future-center regularizer when the static prior is already present remains unclear. This leaves the attribution of the combined regularization gain incomplete, while preserving the evidence for state sharing in Table 10. The description of the final “Reg.” step in Table 3 should explicitly identify the combined losses and distinguish their joint effect from the incremental contribution of L_reg.
W2. Clarity of baseline provenance and evaluation settings. Table 8 explicitly reports a reevaluation of the released SparseWorld-TC-Small checkpoint, but the checkpoint or evaluation differences underlying its results relative to Table 1 are not fully explained. Table 3 is described as a reproduced direct-forecasting baseline, and its architectural and training relationship to SparseWorld-TC could be specified more explicitly. For the 8-second comparison in Table 2, the common test-time extension procedure is stated, but the training horizons of the SparseWorld-TC checkpoint are not specified. These clarifications would help establish the precise scope and comparability of the reported results.
Questions for authors
1.Contribution of the future-center regularizer. Appendix D.3 and Table 11 already discuss the importance of the static-scene prior to the combined regularization gain. Could you additionally report the configuration with (L_{\mathrm{static}}) enabled and (L_{\mathrm{reg}}) disabled, keeping the remaining settings fixed? This would clarify the additional contribution of (L_{\mathrm{reg}}) once the static-scene prior is present. 2.Baseline definitions and evaluation protocol. Table 8 explicitly identifies its SparseWorld-TC-Small results as a reevaluation of the released checkpoint, while Table 3 uses a reproduced direct-forecasting baseline. Could you clarify the checkpoint or evaluation differences underlying the SparseWorld-TC-Small results in Tables 1 and 8, and specify how the Table 3 baseline differs from SparseWorld-TC? For the 8-second comparison in Table 2, please also specify the future horizons supervised during training of the SparseWorld-TC checkpoint. The common test-time extension procedure is already stated, but the checkpoint’s training horizons remain unclear to me. 3.Numerical configuration details. Appendix C provides several useful implementation details, including pretraining information and loss weights. However, I could not locate the numerical values of (\mathbf{b}) in Eq. (12), (v_{\max}) and (\omega_{\max}) in Eq. (14), the static-anchor matching distance, or the occupancy output threshold. Could you provide these values, together with the decoder depth and complete per-layer point schedules for both model sizes, in a compact configuration table or configuration file? For settings inherited from a baseline, identifying the exact configuration would be sufficient. 4.Dependence of current-frame predictions on future trajectory conditions. Appendix E.2 describes identical current-frame forecasts under different future trajectory conditions, while Appendix C.2 describes temporal attention across the (H+1) horizon tokens. Is this invariance guaranteed by the architecture, or is it an observation specific to the illustrated examples? Please clarify the relevant attention mask or inference procedure, if any.
3 comments
Q3. Numerical Configuration Details
Below we list all requested hyperparameters and schedules (shared by both model sizes unless noted). We have also added inline values in Appendices C.2-C.4 and this full summary table in Appendix F (Table 14) of the revised manuscript:
| Parameter / Component | Reference | KIO-World-Small | KIO-World-Large |
|---|---|---|---|
| Number of query anchors | Eq. (2), Sec. 4.1 | ||
| Decoder depth (active motion layers) | Sec. 3.2, App. C.2 | (, rigid) | (, rigid) |
| Per-layer point schedule | Sec. 3.4, App. C.2 | ||
| Final total points per horizon () | Sec. 4.1 | ||
| Feature dim / Point descriptor dim | Tab. 6, App. C.2 | / | / |
| Sampling groups (rigid / motion) points | App. C.2 | ( / ) | ( / ) |
| State bound | Eq. (12) | ||
| Huber threshold in | Eq. (5), App. C.4 | ||
| Static-anchor matching distance | Eq. (13), App. C.4 | (8 static classes) | (8 static classes) |
| Feasibility bounds | Eq. (14), App. C.4 | ||
| Inference occupancy score threshold | App. C.3 | (sigmoid score) | (sigmoid score) |
| Training epochs / batch size / initial LR | Sec. 4.1 | / / | / / |
Q4. Dependence of Current-Frame () Predictions on Future Trajectory Conditions
Thank you for pointing out this distinction between Appendix C.2 and Appendix E.2. Exact invariance of the prediction is not guaranteed by the architecture. Instead, the nearly identical output in Fig. 6 is an empirical behavior. Specifically:
- Direct conditioning at is invariant: At , the ego-pose transform is fixed to the identity with . As a result, both the AdaLN-Zero condition (Eq. 10) and the camera projection in 4D feature sampling do not depend on future ego poses .
- Indirect coupling via unmasked temporal attention: In Appendix C.2, per-query temporal attention across the horizon tokens is unmasked (bidirectional), with no causal mask blocking from attending to . In Fig. 6, all horizons are decoded together in one forward pass, so future trajectory conditions can in principle affect features through
tempo_attn. - Why stays nearly unchanged in practice: Because is directly grounded in current camera observations () and supervised by the same current-frame occupancy ground truth under any future trajectory,
tempo_attnmainly passes context from to future horizons rather than changing geometry based on future maneuvers. Thus, the predictions appear visually indistinguishable in Fig. 6.
We have clarified the unmasked temporal attention setup in Appendix C.2 and changed "identical" to "visually indistinguishable" in Appendix E.2 with a brief explanation.
Before the discussion concludes, we would welcome your thoughts on our response and revised manuscript. In particular, please let us know whether any concerns still require clarification. We remain available to respond during the remaining discussion period. Thank you for your time.
Response to Reviewer ygc8
We thank Reviewer
ygc8for the detailed review and positive assessment of our experiments. We address all questions below. All revisions are marked in blue in the updated manuscript.Q1 & W1. Contribution of Lreg Given Lstatic, and Clarity in Table 3
We trained
KIO-World-Smallfor 70 epochs (keeping HRPR, motion-compensated historical alignment, and Lfeas fixed) with Lstatic enabled and Lreg disabled (Appendix D.3, Table 11):This comparison shows why neither term works alone and both are required together:
+0.2mIoU and+1.0IoU over Lstatic alone). We also updated Section 4.3 and Table 3 to state explicitly that the final "Reg." step includes the combined kinematic losses (Lreg,Lstatic,Lfeas) and refers to Table 11 for their separate effects.Q2 & W2. Baseline Definitions, Checkpoint Provenance, and Evaluation Protocols
SparseWorld-TC-Smallin Table 1 (25.6 / 49.0) vs. Table 8 (25.3 / 48.3):25.6 / 49.0) directly quotes the numbers reported in Du et al. (2026).SparseWorld-TC-Smallcheckpoint with its official evaluation script, which gives25.3mIoU and48.3IoU. We kept the higher paper-reported25.6 / 49.0in Table 1 (favorable to the baseline) and reported the released checkpoint's evaluation in Tables 7-9 (Appendix D.2).Table 3 Baseline vs.
SparseWorld-TC-Small:25.4 / 48.7) is trained in the KIO-World codebase withcenter_offsetrefinement (as in OPUS and SparseWorld-TC), rigid ego-pose sampling, and direct decoding without kinematic losses (Section 4.3).SparseWorld-TCin Table 2 (8-Second Comparison):22.3 / 45.4) directly quotes the officialSparseWorld-TC-Largenumbers reported in Table 2 of Du et al. (2026).SparseWorld-TC-LargeandKIO-World-Largeare supervised strictly on 0-3s during training and directly extended to 0-8s at test time without retraining (jointly decoding all 9 horizons in a single forward pass), as now explicitly stated in Section 4.2 and Table 2.