Video LLM Motion-Direction Auxiliary Branch Diagram
Overview
A two-band diagram of a video LLM with an auxiliary motion-direction objective. It is an editable PowerPoint diagram adapted from the original paper figure under CC BY 4.0.
What the diagram shows
The top band shows the training-and-inference path: input video frames → frozen vision encoder → trainable projector → visual tokens F ∈ ℝ^{T×N×D} plus text tokens from a question prompt → large language model → output text, supervised by L_LM. The bottom band is a training-only branch that is removed at inference: feature deltas between F_t and F_{t+1}, pooling Pool_N(·) into a motion descriptor δ_t ∈ ℝ^D, a motion prediction head, and an L_MVP loss between the predicted and target motion dials. A side panel gives the total objective L_total = L_LM + λL_MVP and the normalized-displacement motion target m_t, with a trainable / frozen / training-only legend. Every box, arrow, label and math term is a native shape you can recolor, resize and relabel for your own slides, posters or papers.
File format
The download contains .pptx (1 slide, 13.33 × 4.71 in; opens in PowerPoint, Keynote, WPS and LibreOffice Impress; Google Slides works too, but complex freeforms may shift slightly), .svg and .drawio. Includes preview.png (full resolution) and a 1200 px web preview.
Shapes and editing
The file has 103 native editable shapes: 26 auto-shapes, 43 text boxes, 12 freeforms, 22 lines/connectors. No embedded images. Everything is vector. You can edit all text and math (𝐅 ∈ ℝ^{T×N×D}, 𝐅_t, 𝐅_{t+1}, Pool_N(·), δ_t ∈ ℝ^D, L_MVP, L_total, the m_t fraction with norm bars), the stacked input-frame cards with the t → t+1 motion arrow, encoder / projector / LLM blocks, visual and text token pills, the question bubble, both motion dials, the feature-delta bar, dashed training-only routes, the legend and native flame / snowflake icons.
Embedded images and fonts
This file has no embedded images; everything is vector. Text uses Times New Roman, which is preinstalled on Windows and macOS (LibreOffice uses the metric-compatible Liberation Serif). The double-struck ℝ uses Cambria Math, which ships with Microsoft Office; other apps substitute a math font. Sub- and superscripts are real formatted text runs, not equation objects. The SVG keeps text as text; the draw.io file is a shape-by-shape conversion and may need small nudges. This is a faithful redraw for reuse, not the authors' original source file.
Source & license Adapted from Figure 6 of Which Way Did It Move? Diagnosing and Overcoming Directional Motion Blindness in Video-LLMs by Jongseo Lee, Hyuntak Lee, Sunghun Kim, Sooa Kim, Jihoon Chung, Jinwoo Choi (arXiv:2605.22823, https://arxiv.org/abs/2605.22823). Licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). Redrawn as editable shapes; changes made by 叫我Alonzo就好了/alonzolab. Not endorsed by the original authors. Attribution to the original paper is required when you publish or share the figure; the exact credit line is in LICENSE-ATTRIBUTION.txt inside each download.
Questions or feedback? Email [email protected]