- Rating
- 3(+1)
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper proposes 4DThinker, a framework for dynamic spatial reasoning from monocular video. It introduces a scalable data pipeline for synthesizing 4D reasoning data from raw videos. Dynamic-Imagery Fine-Tuning (DIFT) combines text-token cross-entropy with a cosine-similarity objective that aligns predicted hidden states with pooled visual embeddings of the highlighted frames. 4D Reinforcement Learning (4DRL) improves the reasoning capability using correctness and format rewards. The experiments show that Qwen3-VL-32B/InternVL3.5-38B based 4DThinker outperforming the compared baselines on DSR-Bench and Dyn-Bench.
Critical strengths
-
Consistent improvements across multiple benchmarks and backbones. The experiments demonstrate performance gains on DSR-Bench and Dyn-Bench using Qwen2.5-VL-3B/7B, Qwen3-VL-8B/32B, and InternVL3.5-8B/38B.
-
Informative ablation studies. The ablation studies through comparisons across training strategies, loss terms, and reward components, which support the contributions of the major components of 4DThinker.
Critical weaknesses
-
Limited methodological novelty. DIFT closely resembles the latent visual grounding approach introduced in Machine Mental Imagery. Machine Mental Imagery also explicitly optimizes only text-token probabilities while allowing gradients to propagate through latent states. LVR addresses the absence of discrete token distributions at continuous latent positions through a GRPO variant that computes the policy-gradient objective solely over text-generation tokens, which is similar with 4DRL. The paper should clarify the substantive differences from these methods and explain what makes 4DRL specifically suited to 4D reasoning.
-
The comparison with DSR Suite-Model does not control in backbone size. Although Table 1 shows that Qwen3-VL-32B based 4DThinker outperforms DSR Suite-Model (62.0 vs 58.9), the latter uses the substantially smaller Qwen2.5-VL-7B backbone. When using the same Qwen2.5-VL-7B backbone, 4DThinker achieves a score of 47.1, substantially below DSR Suite-Model’s 58.9. Therefore, the results do not establish a performance advantage of the proposed method over the baselines.
Questions for authors
Please see the weaknesses above.
5 comments
Response to Reviewer 9mPE [2/2]
W2: Controlled Backbone-Scale (7B) and Matched-Dataset Comparison with DSR Suite-Model
Thank you for pointing out this fairness issue. In Sec. 4.1 of the revised manuscript, we will revise the cross-backbone phrasing and combine our additional evaluations from Reviewer dFA5 (W1/Q1) and Reviewer gaQe (W1, W2) to present controlled comparisons under the same Qwen2.5-VL-7B backbone across all five dynamic spatial reasoning benchmarks and their overall mean (Avg.):
| Controlled Group | Model | Backbone | DSR Training Data | External Geometric Modules at Inference | DSR-Bench | Dyn-Bench | VLM4D | DSI-Bench | STI-Bench | Avg. |
|---|---|---|---|---|---|---|---|---|---|---|
| Same 7B | DSR Suite-Model | Qwen2.5-VL-7B | Official full 50K | GSM ( + SAM2 + Orient) | 58.9 | 60.8 | 48.3 | 38.6 | 35.2 | 48.4 |
| Same 7B | 4DThinker (main paper) | Qwen2.5-VL-7B | 37K subset | None | 47.1 | 65.9 | 54.2 | 46.8 | 44.8 | 51.8 |
| Same 7B | 4DThinker (full DSR-Train 50K) | Qwen2.5-VL-7B | Official full 50K | None | 60.8 | 66.8 | 55.1 | 47.6 | 45.7 | 55.2 |
This controlled comparison clarifies the performance of 4DThinker at matched backbone scale:
Under the shared Qwen2.5-VL-7B backbone, 4DThinker already leads in five-benchmark average and surpasses DSR Suite-Model on all five benchmarks once aligned to the full 50K training set: The 7B gap on DSR-Bench in the main paper (47.1 vs. 58.9) stems primarily from differences in training data volume and external modules. DSR Suite-Model trains on the full DSR-Train 50K set and attaches three external vision/3D models (, Grounded-SAM2, and Orient Anything) at inference to inject explicit geometric tokens. By contrast, 4DThinker-7B in the main paper uses a 37K subset with sparse outcome rewards and zero external modules at inference. Even under the main-paper setting, 4DThinker-7B outperforms DSR Suite-Model by 5.1–9.6 points across all four cross-distribution benchmarks (Dyn-Bench, VLM4D, DSI-Bench, and STI-Bench), reaching a five-benchmark average of 51.8 (above DSR Suite-Model at 48.4). When aligned to the exact same official full DSR-Train 50K set (as in Reviewer dFA5 W1/Q1, using DSR-Train 50K in both stages), module-free 4DThinker-7B reaches 60.8 on DSR-Bench (exceeding DSR Suite-Model at 58.9) and raises the five-benchmark average to 55.2 (+6.8 points higher).
In summary, under a controlled comparison with the same 7B backbone + the same 50K training set and zero external geometric modules, 4DThinker outperforms DSR Suite-Model both on DSR-Bench and in average accuracy across all five dynamic spatial benchmarks (7B: 55.2 vs. 48.4). We will add these controlled comparisons to the revised manuscript, and thank you again for this constructive feedback.
Sincere Thanks and Follow-up to Reviewer 9mPE
Dear Reviewer 9mPE,
Thank you again for taking the time to review our manuscript and share your feedback. We have responded to your comments and revised the manuscript accordingly. We hope that our responses and revisions address your concerns. If any questions or concerns remain, we would be happy to address them during the discussion period.
Best regards,
The Authors
Response to Reviewer 9mPE [1/2]
We sincerely thank the reviewer for the constructive comments and valuable suggestions. We address each point below and will revise the manuscript accordingly.
W1: Substantive Differences from
Machine Mental Imagery (Mirage)andLVR, and Why4DRLSuits 4D ReasoningThank you for highlighting these two representative prior works. We cite both
Mirage(Yang et al., 2025c) andLVR(Li et al., 2025a) in Sec. 2.1, and will add a dedicated comparison paragraph in the revised manuscript clarifying our contributions across five dimensions: problem space, supervision construction, annotation dependency, temporal structure, and RL dynamics:think–imagery–answerrewards for compound-motion explorationBeyond applying
MirageorLVRto video inputs,4DThinkerresolves two technical challenges that static latent reasoning does not address:DIFT):LVRlatents zoom into local regions of an already observed single image, whileMiragerelies on manually curated static auxiliary views. Neither represents an object accelerating or turning while the camera translates and rotates. In dynamic video reasoning, collapsing supervision to a single static timestamp (Last-frame DIFTin Table 6(a)) yields only 26.7 accuracy, matching text-onlyCoT SFT(26.8). Introducing temporally ordered, mask-grounded sequences enables the model to relate the camera reference frame (static landmark displacement) to object motion trajectories in latent space, raising accuracy to 31.1. Under the sharedQwen2.5-VL-7Bbackbone (Table 12),4DThinkerreaches 47.1 / 65.9 onDSR-Bench/Dyn-Bench, outperforming the static latent-reasoning baselineLVR(33.5 / 58.4) by a wide margin.4DRL): As you noted, computing policy gradients only on text tokens (since continuous latent positions lack a discrete vocabulary distribution) shares the same formal principle asLVR. In4DThinker, however, latent states are not single-step static feature extractions. They form a directed temporal trajectory generated recursively across multiple timestamps via Eq. (10), and4DRLmust compose camera and object motion reasoning autonomously on compound-motion data (DSR-Train) without intermediate visual annotations. Without masking discrete losses at latent positions (vanilla GRPOin Table 6(c)), high-variance token-level gradients backpropagate along the recursive chain across multi-step latent states. Cosine similarity to ground-truth dynamic overlays then drops from 0.63 (DIFTinitialization) to 0.51, weakening cross-video causal grounding. By contrast,4DRLmaintains cosine similarity at 0.61 and outperformsvanilla GRPOonDSR-Bench(34.2 vs. 32.5, and combining Rfmt to verify the temporally interleavedthink–imagery–answerstructure also improves over 33.4 without the format reward).In summary,
MirageandLVRtarget localized static-image representations with manual annotations, whereas4DThinkerenables annotation-free cross-timestamp simulation of coupled camera-object motion in monocular video. Under the same 7B backbone,4DThinkerimproves overLVRby +13.6 / +7.5 points, and our ablations and cosine tracking verify that multi-step temporal ordering and recursive gradient isolation are both necessary in 4D settings.