AI Paper Atlas

Research map / Vision-language and multimodal / Robot action and navigation

FGPrompt: Fine-grained Goal Prompting for Image-goal Navigation

Xinyu Sun, Peihao Chen, Jugang Fan, Jian Chen, Thomas H. Li, Mingkui Tan

NeurIPS 2023 · Vision-language and multimodal › Robot action and navigation · approach: Pretraining and data augmentation for VLN · 4 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Condition observation features on fine-grained goal features with FiLM layersFGPrompt-MF · MP3D cross-domainSR: 14.6% -> 77.6%Improved
Replace late fusion with channel concatenation and joint modeling (ablation)ImageNav navigation policy · Gibson ImageNavSR: 13.0% -> 78.9%Improved
Add keypoint-matching fusion to late fusion (ablation)ImageNav navigation policy · Gibson ImageNavSR: 13.0% -> 41.6%Improved
Preserve spatial resolution rather than average-pooling goal activations for FiLM mapping (ablation)FGPrompt-MF · Gibson ImageNavSR: 32.0% -> 77.3%Improved
Fuse at early ResNet blocks instead of increasing fusion depth to 4 blocks (ablation)FGPrompt-MF · Gibson ImageNavSR: 71.4% -> 77.3%Improved
Concatenate goal and observation images along channels instead of stacking them for 3D convolution (ablation)FGPrompt-EF · Gibson ImageNavSR: 20.5% -> 78.9%Improved
Jointly model goal and observation images rather than share encoder parameters (ablation)FGPrompt-EF · Gibson ImageNavSR: 14.6% -> 78.9%Improved
Use mid fusion rather than no fusion under randomly varied goal-camera parameters (ablation)ImageNav navigation policy · Augmented camera-parameter ImageNavSR: 12.1% -> 70.2%Improved
Use mid fusion rather than early fusion (ablation)ImageNav navigation policy · Augmented camera-parameter ImageNavSR: 64.6% -> 70.2%Improved
Use mid fusion rather than early fusion (ablation)ImageNav navigation policy · Instance image navigationSR: 3.4% -> 9.9%Improved
Add early fusion between the unshuffled image and current observationResNet18+IL rearrangement baseline · AI2THOR 1-Phase RearrangementSuccess: 1.89 -> 10.2 (+439%)Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy