Temporal Event Proposal Rank-Calibration Diagram

Overview

A four-panel diagram of calibrating temporal event proposals in videos. It is an editable PowerPoint diagram adapted from the original paper figure under CC BY 4.0.

What the diagram shows

(a) Query-based event candidate generation: a video database icon and a film strip feed two frozen video encoders (A and B, snowflake badges), whose feature-token rows go into a Transformer decoder (learnable queries, a self-attention block with curved arcs and an FFN grid, a query head), giving two proposal timelines of (segment, label, score) bars. (b) Proposal rank calibration in four numbered steps: 1 cross-representation consensus calibration (tIOU matching between the two proposal sets, a trainable α / β weighted sum, proposal bars before and after); 2 localization quality estimation (proposal stack → MLP → quality score q^loc_i); 3 proposal redundant suppression (a proposal graph → relation encoder → relation-suppression score d_i); 4 post processing (temporal box voting and multi-class NMS); both scores join a score calculation card that re-ranks the proposals into the final detection, with a trained / frozen legend. (c) One-stage loc-cls detection protocol: video frames → encoder → feature column → loc-cls detector with a tIOU-by-class chart → comprehensive evaluation and final output events. (d) Two-stage loc-first protocol: encoder → loc-only detector with an event / background chart → loc-evaluation → trainable projector → frozen LLM (classify / caption) → comprehensive evaluation and final output. 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, 13.84 × 9.47 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 718 native editable shapes: 404 auto-shapes, 105 text boxes, 22 freeforms, 187 lines/connectors (10 small groups). No embedded images. Everything is vector. You can edit all text and math (α, β, Σ, q^loc_i, d_i), the film strips and database icon, both video-encoder boxes, every feature-token cell, the Transformer decoder with its query columns, attention arcs and FFN grid, all proposal bars and timelines, the tIOU matching links, the weighted-sum block, the before / after proposal columns, the MLP and relation-encoder node graphs, the proposal graph, the score calculation card with its re-rank cross, the voting and NMS bars, the flame and snowflake badges, the detector charts, the encoder trapezoids, the projector and LLM bars, the evaluation cards, the numbered step badges, all arrows and every panel.

Embedded images and fonts

This file has no embedded images; everything is vector. Every video frame (the film strip at the top and the two vertical strips in the protocol panels) is a simplified flat vector placeholder of sky, ridge, hill and ground blocks with no people in it, not the paper's video frames; paste your own frames over them if you need real samples. All text, body and formulas alike, uses Times New Roman, which is preinstalled on Windows and macOS (LibreOffice substitutes the metric-compatible Liberation Serif). 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.