@@ -147,16 +147,12 @@ <h2 class="title is-3">Abstract</h2>
147147 < h2 class ="title is-3 has-text-centered "> Reconstruct images from feature space</ h2 >
148148 < div class ="content ">
149149 < div class ="has-text-centered ">
150- < figure class ="image is-16by9 mb-6 ">
151- < img src ="./static/images/features_reconstruction.drawio.png " alt ="features_reconstruction " class ="is-rounded ">
152- </ figure >
150+ < img src ="./static/images/features_reconstruction.drawio.png " alt ="features_reconstruction " class ="is-rounded ">
153151 < p class ="caption has-text-centered mb-6 ">
154152 < b > Figure 1. Image reconstructor training.</ b > For pretrained model we train a reconstructor model that
155153 restores the image from the feature space.
156154 </ p >
157- < figure class ="image is-16by9 mb-6 ">
158- < img src ="./static/images/reconstruction_metrics.jpg " alt ="reconstruction_metrics " class ="is-rounded ">
159- </ figure >
155+ < img src ="./static/images/reconstruction_metrics.jpg " alt ="reconstruction_metrics " class ="is-rounded ">
160156 < p class ="caption has-text-centered mb-6 ">
161157 < b > Figure 2. Reconstruction Metrics.</ b > We show the results of the reconstruction for SigLip and SigLip2
162158 for different image resultions.
@@ -185,17 +181,13 @@ <h2 class="title is-3 has-text-centered">Feature-space transformations. Q matrix
185181 < div class ="content ">
186182 < div class ="columns is-centered ">
187183 < div class ="column is-half ">
188- < figure class ="image is-16by9 mb-6 ">
189- < img src ="./static/images/features_reconstruction_manipulation_train_Q.drawio.png " alt ="features_reconstruction_manipulation_train_Q " class ="is-rounded ">
190- </ figure >
184+ < img src ="./static/images/features_reconstruction_manipulation_train_Q.drawio.png " alt ="features_reconstruction_manipulation_train_Q " class ="is-rounded ">
191185 < p class ="caption has-text-centered mb-6 ">
192186 < b > Figure 3. Feature-space transformations. Q matrix Calculation.</ b > We then calculate Q matrix for feature-space manupulation.
193187 </ p >
194188 </ div >
195189 < div class ="column is-half ">
196- < figure class ="image is-16by9 mb-6 ">
197- < img src ="./static/images/features_reconstruction_manipulation_eval_Q.drawio.png " alt ="features_reconstruction_manipulation_eval_Q " class ="is-rounded ">
198- </ figure >
190+ < img src ="./static/images/features_reconstruction_manipulation_eval_Q.drawio.png " alt ="features_reconstruction_manipulation_eval_Q " class ="is-rounded ">
199191 < p class ="caption has-text-centered mb-6 ">
200192 < b > Figure 4. Feature-space transformations. Q matrix Application.</ b > After Q matrix is calculated, we apply it to the feature space. For each patch embedding.
201193 </ p >
@@ -215,17 +207,13 @@ <h2 class="title is-3 has-text-centered">Feature-space transformations. Color Sw
215207 < div class ="content ">
216208 < div class ="columns is-centered ">
217209 < div class ="column is-half ">
218- < figure class ="image is-16by9 mb-6 ">
219- < img src ="./static/images/rb_swap.png " alt ="rb_swap " class ="is-rounded ">
220- </ figure >
210+ < img src ="./static/images/rb_swap.png " alt ="rb_swap " class ="is-rounded ">
221211 < p class ="caption has-text-centered mb-6 ">
222212 < b > Figure 5. Red-blue channel swap samples.</ b >
223213 </ p >
224214 </ div >
225215 < div class ="column is-half ">
226- < figure class ="image is-16by9 mb-6 ">
227- < img src ="./static/images/color_swap_all_eigen_values.png " alt ="color_swap_all_eigen_values " class ="is-rounded ">
228- </ figure >
216+ < img src ="./static/images/color_swap_all_eigen_values.png " alt ="color_swap_all_eigen_values " class ="is-rounded ">
229217 < p class ="caption has-text-centered mb-6 ">
230218 < b > Figure 6. 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.
231219 </ p >
@@ -245,17 +233,13 @@ <h2 class="title is-3 has-text-centered">Feature-space transformations. Blue Cha
245233 < div class ="content ">
246234 < div class ="columns is-centered ">
247235 < div class ="column is-half ">
248- < figure class ="image is-16by9 mb-6 ">
249- < img src ="./static/images/b_suppression_all_transformations.png " alt ="b_suppression_all_transformations " class ="is-rounded ">
250- </ figure >
236+ < img src ="./static/images/b_suppression_all_transformations.png " alt ="b_suppression_all_transformations " class ="is-rounded ">
251237 < p class ="caption has-text-centered mb-6 ">
252238 < b > Figure 7. Blue channel suppression samples.</ b >
253239 </ p >
254240 </ div >
255241 < div class ="column is-half ">
256- < figure class ="image is-16by9 mb-6 ">
257- < img src ="./static/images/b_suppression_all_eigen_values.png " alt ="b_suppression_all_eigen_values " class ="is-rounded ">
258- </ figure >
242+ < img src ="./static/images/b_suppression_all_eigen_values.png " alt ="b_suppression_all_eigen_values " class ="is-rounded ">
259243 < p class ="caption has-text-centered mb-6 ">
260244 < b > Figure 8. Eigenvalues for blue channel suppression matrix.</ b >
261245 </ p >
@@ -274,9 +258,7 @@ <h2 class="title is-3 has-text-centered">Feature-space transformations. Blue Cha
274258 < h2 class ="title is-3 has-text-centered "> Feature-space transformations. Colorization.</ h2 >
275259 < div class ="content ">
276260 < div class ="has-text-centered ">
277- < figure class ="image is-16by9 mb-6 ">
278- < img src ="./static/images/colorized_examples.png " alt ="colorization_all_transformations " class ="is-rounded ">
279- </ figure >
261+ < img src ="./static/images/colorized_examples.png " alt ="colorization_all_transformations " class ="is-rounded ">
280262 < p class ="caption has-text-centered mb-6 ">
281263 < b > Figure 9. Colorization samples.</ b >
282264 </ p >
0 commit comments