Figureure 2 · Progressive simplification of cycle consistency to Geometric ReciprociFigure 2. Progressive simplification of cycle consistency to Geometric Reciprocity. (i) Inpainted regions in $\hat { I } _ { L }$ do not affec $\hat { I } _ { R } ^ { \mathrm { r e c o n } }$ allowing us to skip left view inpainting (marked with ×) and directly use $\tilde { I } _ { L }$ . (ii) Right-to-left warping transfers disparity from $d _ { R }$ to $\tilde { I } _ { L }$ , allowing us to skip left view disparity estimation and directly reuse $\tilde { d } _ { L }$ . (iii) Transferred disparity ensures perfect round-trips for all validly warped pixels, enabling analytical computation of $M _ { \mathrm { d i s } } ^ { L }$ as pixels lost during right-to-left warping and eliminating all warping operations (marked with ×). The final result reveals that $M _ { \mathrm { d i s } } ^ { L R } = M _ { \mathrm { l o s t } } ^ { R L }$ can be computed directly from $( I _ { R } , d _ { R } )$ alone.这张图来自论文 PDF 的结构化抽取。当前用于辅助理解 Geometric Reciprocity 的方法或实验，请结合正文精读段落一起看。Figure A2Figure A2. Performance of SDXL Inpainting on different mask patterns. The model handles large contiguous masks well (General Inpainting) but struggles with thin scattered disocclusion masks along object boundaries (Stereo Inpainting).这张图/表用于判断 Geometric Reciprocity 的实验收益来自哪里。重点看替换、消融或跨模型设置下趋势是否一致，而不是只看单个最高分。
