@@ -143,11 +143,11 @@ <h2 class="title is-3">Abstract</h2>
143143 < h2 class ="title is-3 " style ="white-space: nowrap; "> Interpretability metric</ h2 >
144144 < div class ="content has-text-justified ">
145145 < div style ="text-align: center; ">
146- < img src ="./static/images/features_reconstruction.drawio.png " alt ="Data Mixture " width ="900 ">
146+ < img src ="./static/images/features_reconstruction.drawio.png " alt ="features_reconstruction " width ="900 ">
147147 < p class ="caption " style ="width: 100%; text-align: center; ">
148148 < b > Figure 1. Image reconstructor training.</ b > For pretrained model we train a reconstructor model that restores the image from the feature space.
149149 </ p >
150- < img src ="./static/images/reconstruction_results.png " alt ="Data Mixture " width ="900 ">
150+ < img src ="./static/images/reconstruction_results.png " alt ="reconstruction_results " width ="900 ">
151151 < p class ="caption " style ="width: 100%; text-align: center; ">
152152 < b > Figure 2. Reconstruction results.</ b > We show the results of the reconstruction for SigLip and SigLip2 for different image resultions.
153153 </ p >
@@ -170,15 +170,15 @@ <h2 class="title is-3" style="white-space: nowrap;">Interpretability metric</h2>
170170 < h2 class ="title is-3 " style ="white-space: nowrap; "> Feature-space transformations</ h2 >
171171 < div class ="content has-text-justified ">
172172 < div style ="text-align: center; ">
173- < img src ="./static/images/features_reconstruction_manipulation.drawio.png " alt ="Data Mixture " width ="900 ">
173+ < img src ="./static/images/features_reconstruction_manipulation.drawio.png " alt ="features_reconstruction_manipulation " width ="900 ">
174174 < p class ="caption " style ="width: 100%; text-align: center; ">
175175 < b > Figure 3. Feature-space transformations.</ b > We then calculate Q matrix for feature-space manupulation.
176176 </ p >
177- < img src ="./static/images/rb_swap.png " alt ="Data Mixture " width ="900 ">
177+ < img src ="./static/images/rb_swap.png " alt ="rb_swap " width ="900 ">
178178 < p class ="caption " style ="width: 100%; text-align: center; ">
179179 < b > Figure 4. Red-blue channel swap samples.</ b > We then calculate Q matrix for feature-space manupulation.
180180 </ p >
181- < img src ="./static/images/rb_swap_eigenvalues.png " alt ="Data Mixture " width ="900 ">
181+ < img src ="./static/images/rb_swap_eigenvalues.png " alt ="rb_swap_eigenvalues " width ="900 ">
182182 < p class ="caption " style ="width: 100%; text-align: center; ">
183183 < b > Figure 5. Eigenvalues for red-blue channel swap matrix.</ b > Majority of eigenvalues are close to 1, which means that the transformation is close to an identity matrix. While the other cluster of eigenvalues are close to -1, which means that for these channels direction is changed to the opposite.
184184 </ p >
0 commit comments