- Rating
- 3
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 1(−1)
- Soundness
- 2
Read review
Paper summary
This paper proposes an object-centric streaming video memory mechanism. It indexes raw KV tokens using semantic slots and maintains cross-segment memory through entity association, snapshot retrieval, and fixed-budget assembly. Built upon the Wan2.1-1.3B model, this approach enhances dynamic motion and improves text-video alignment in the latter portions of long-turn videos and multi-prompt narratives.
Critical strengths
Provided solution is well-designed: Slots handle semantic addressing, and attention reads the original KVs, avoiding compressing all visual details into low-dimensional slot representations. Causal read-write order, entity lifecycle, and retrieval budget have been clarified clearly in Appendix B.
Critical weaknesses
- The experimental results are not particularly robust.: Text alignment and FG-ARI/mBO metrics cannot measure post-departure identity recovery. Since dormant is disabled by default (shwon in Appendices B.4–B.5), returning objects spawn new entities. Thus, identity switching and memory duplication during repeated re-entry remain unverified. Moreover, each layer allows up to 8192 entity snapshot tokens besides 2688 active tokens (App. B). Table 4 fails to clarify if Top-k baselines match this background storage. thus, the "structure over capacity" attribution requires strictly controlled total memory.
- Figure 2 is not particularly clear.: the font size is a little small, the process relationships are not clearly conveyed, and the core design elements do not stand out.
Questions for authors
In re-entry tests, how do object count, similarity, and absence duration affect identity recovery and duplication rates? Can the solution maintain identity consistency? How do results shift when re-identification is enabled and the bank hits capacity?
4 comments
Part 2/2: W1 & Q1, parts 3-4, and W2
3. Effect of Object Count, Similarity, Enabling Re-Identification, and Hitting Bank Capacity (Q1).
Object count and inter-object similarity. Each semantic layer tracks up to entities and matches per-chunk slots via gated cosine similarity ( threshold for ACTIVE/TENTATIVE) and Hungarian matching (App. B.4). Higher object count increases slot competition, while high visual similarity across distinct objects is disambiguated by one-to-one Hungarian assignment and relative RoPE phase cues (App. D.3), though visually indistinguishable objects can either share an entity record or spawn separate TENTATIVE entries if below the gate.
Shift when re-identification is enabled. As shown in the table above, enabling the DORMANT gate improves short-gap recovery at ( vs. , ), where RoPE phase shift is small, but slightly lowers consistency over longer gaps ( at , at , and at ) due to noisy long-horizon descriptor matching updating stored snapshots.
Shift when the bank hits capacity. Each semantic layer caps storage at entities ( snapshots each). Once full, DORMANT and low-quality TENTATIVE entries are evicted by lowest retention score (combining quality, match count, and dormancy age), whereas ACTIVE entities are never evicted; if no evictable entry exists, unmatched slots are discarded rather than overwriting supported identities (App. B.4). We will include these quantitative comparisons and analyses in the revised manuscript.
4. Memory Accounting and "Structure over Capacity" Attribution (W1)
Thank you for catching this distinction. Per semantic layer, Table 4 matches the attention-visible budget at active tokens (), while SlotMemory additionally maintains up to entity snapshot tokens () in background storage. Retrieved snapshots occupy slots inside the -token attention budget rather than expanding self-attention context.
Three findings in the paper support that our gains stem from entity-level addressing rather than raw token capacity:
Capacity saturation vs. unstructured decay (Figure 5). Scaling SlotMemory's capacity across blocks—specifically doubling from blocks ( tokens) to blocks ( tokens)—produces minimal performance change, with the -block default achieving the highest overall VBench quality. In contrast, unstructured memory designs (frame-sink caching and NAM) lose roughly twice as much text alignment by the final segment (), demonstrating that the bottleneck is entity-level routing rather than stored token count.
Isolating entity association under a fixed -token attention budget (Table 4). Without cross-chunk identity tracking, both untyped Top-k KV Memory and + Slot (no entity bank) suffer severe late-rollout Text Drift ( and ). Adding the entity bank (Ours (full)) cuts Text Drift to and raises mBO from to while injecting the exact same active tokens into self-attention.
Strictly fixed total storage ( tokens, Table 5 & Appendix Table D.6). Both ablations hold active and background capacities strictly identical, yet varying slot regularization or identity-key representation shifts Dynamic Degree ( to ), Subject Consistency ( to ), and alignment drift ( to ).
W2: Clarity and Visual Hierarchy of Figure 2
We have redesigned Figure 2 with a larger font size, clearer directional connections conveying the process relationships between stages, and improved visual hierarchy.
Thanks for the authors’ response, which has addressed most of my concerns about the experiment. However, I remain concerned about the clarity of Figure 2. The inconsistent font sizes, crowded layout, and overlaps between text and boxes make the figure difficult to read. As the paper’s critical overview figure, it should provide an intuitive explanation of the method’s main components and their interactions to help readers understand the proposed method. In its current form, it is dominated by dense textual annotations, making the overall workflow difficult to follow. I therefore lower my presentation score to 1. Nevertheless, the work appears to reflect substantial effort, and the response has resolved most of my experimental concerns. I will retain my original overall rating of Borderline Accept.
We appreciate the reviewer’s constructive feedback on Figure 2. We apologize that the original figure was cluttered and difficult to follow. To address this, we have thoroughly revised Figure 2 in the updated manuscript with the following improvements:
- Font and layout: We standardized and enlarged font sizes and expanded the margins around the text boxes to eliminate overlaps.
- Content simplification: We have removed dense textual descriptions in favor of intuitive visual schematics.
- Workflow clarity: We have streamlined the arrows and layout to clearly correspond to the top “read-write-update” cycle from the bottom component interactions (slot compression, entity bank, and assembly).
Thank you again for helping us improve the presentation quality of the paper.
Part 1/2: Response overview and W1 & Q1, parts 1-2
We sincerely thank Reviewer RW8m for the careful analysis and constructive feedback, which prompted us to verify our
DORMANTretrieval policy and clarify our memory accounting. In response, we have added a controlled occlusion–reappearance evaluation sweeping absence durations (k∈{9,18,36,54}) and comparing default retrieval against the 0.78DORMANTgate alongside explicit active (2,688-token) vs. background (8,192-token) memory accounting (W1 & Q1), as well as a redesigned Figure 2 with larger typography and clearer process flows (W2).W1 & Q1: Post-Departure Identity Recovery, Re-Identification, and Memory Capacity Behavior
1. Direct Evaluation of Post-Departure Identity Recovery and Absence Duration (W1 & Q1).
While full-sequence Text Align, FG-ARI, and mBO evaluate global prompt compliance and object decomposition, we directly benchmark post-departure identity recovery and identity switching across varying absence durations using a RememBench-derived sequence: Appear(9s)→Close(3s)→Hidden(kchunks)→Reappear(6s) (Tc=3 latent frames, or 0.75s per chunk). During
Hidden, the target object remains fully sealed inside a container across k∈{9,18,36,54} chunks (6.75s, 13.5s, 27.0s, and 40.5s), surpassing the 3.0s (12-latent-frame) local cache horizon. We report MBench Object Texture Consistency (cross-interval similarity of object-level appearance attributes acrossAppearandReappear) alongside a variant enabling the 0.78DORMANTre-identification gate:SlotMemory (default)outperforms MemFlow across all absence durations by +9.62 to +26.57 points without degrading through k=54 (40.5s), supporting reliable visual identity recovery after prolonged absence from the local window. The upward trend with k stems from how MBench Object Texture Consistency validates concealment by discarding rollouts where the object remains visible duringHidden. Under brief occlusion (k=9), lingering visual cues fromAppearandCloseleave the object partially unoccluded more often; longer occlusion spans afford more chunks under the container prompt to completely remove the object from the frame, while SlotMemory retains itsDORMANTK/V snapshots intact for retrieval onceReappearbegins.2. Preventing Snapshot Corruption and Memory Duplication Under Default Retrieval (W1).
Excluding
DORMANTentities from association by default (App. B.4–B.5) is a deliberate safeguard rather than an unverified gap:Because Lcon supervises slot transitions only across adjacent chunks and cached RoPE phases induce cosine-similarity drift over extended intervals (
App. D.3), forcibly re-linking slots after long absences can bind an unrelated object to an existing record and overwrite its stored K/V snapshots. LeavingDORMANTentries outside online matching lets the persistent retrieval step inject clean pre-departure snapshots directly into self-attention without altering stored entity ownership.Even when a returning object initializes a new
TENTATIVErecord, self-attention reads a strictly fixed-size bank Blc of at most Nbudget=6 blocks (42 groups / 2,688 tokens) per semantic layer (App. B.1). During bank assembly (App. B.5), at most one snapshot per entity is selected, and candidate history groups whose mean-pooled key cosine similarity exceeds 0.98 with a retrieved snapshot are deduplicated and dropped, preventing redundant K/V tokens from entering attention during repeated re-entry.