Skip to content

Commit 97d8e4e

Browse files
committed
Add papers
1 parent f2f8841 commit 97d8e4e

21 files changed

+61
-102
lines changed

_projects/1_project.md

Lines changed: 8 additions & 18 deletions
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22
layout: page
33
title: Deep Hough Voting
44
description: 3D object detection
5-
img: assets/img/project/Npm3d/project_deephoughvoting.png
5+
img: assets/img/project/deep_hough_votting/project_deephoughvoting.png
66
importance: 1
77
category: master
88
---
@@ -13,20 +13,10 @@ category: master
1313

1414
**Results:** You can find below the results obtained when running the code on the github repository. Details on the training and on the implementation can be found in the following <a href="/assets/pdf/Report_Deephoughvoting.pdf"> pdf. </a>
1515

16-
<div class="row">
17-
<div class="col-sm mt-3 mt-md-0">
18-
{% include figure.html path="assets/img/project/Npm3d/project_deephoughvoting_results.jpg" title="example image" class="img-fluid rounded z-depth-1" %}
19-
</div>
20-
</div>
21-
<div class="caption">
22-
Point cloud, ground-truth and prediction on some scans of the 3rd testing batch of ScannetV2.
23-
</div>
24-
25-
<div class="row">
26-
<div class="col-sm mt-4 mt-md-0">
27-
{% include figure.html path="assets/img/project/Npm3d/project_deephoughvoting_results_3.jpg" title="example image" class="img-fluid rounded z-depth-1" %}
28-
</div>
29-
</div>
30-
<div class="caption">
31-
Point cloud, ground-truth and prediction on some scans of the 10th testing batch of Sun RGB-D v2.
32-
</div>
16+
<div class="text-center">
17+
<div class="mt-3 mt-md-0">{% include figure.html path="/assets/img/project/deep_hough_votting/project_deephoughvoting_results.png" title="example image" class="img-fluid rounded z-depth-1" %}</div>
18+
<div class="caption">Point cloud, ground-truth and prediction on some scans of the 3rd testing batch of ScannetV2.</div>
19+
20+
<div class="mt-4 mt-md-0">{% include figure.html path="/assets/img/project/deep_hough_votting/project_deephoughvoting_results_3.png" title="example image" class="img-fluid rounded z-depth-1" %}</div>
21+
<div class="caption">Point cloud, ground-truth and prediction on some scans of the 10th testing batch of Sun RGB-D v2.</div>
22+
</div>

_projects/2_project.md

Lines changed: 9 additions & 28 deletions
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22
layout: page
33
title: Omnimatte
44
description: Video inpainting
5-
img: assets/img/project/DeepL/project_omnimatte.png
5+
img: assets/img/project/omnimatte/project_omnimatte.png
66
importance: 1
77
category: master
88
---
@@ -12,32 +12,13 @@ category: master
1212
**Summary:** Video inpainting is the task of reconstructing missing pixels in a video. It is an important problem in computer vision and an essential feature in many imaging and graphic applications, e.g. object removal, image restoration, manipulation, retargeting, image composition and rendering. While image inpainting is an almost solved problem, video inpainting is more difficult to solve as approaches are often unable to maintain the sharpness of the edges and create blurry effects while being unable to remove the correlated effects of an object. In addition, some suffer from temporal coherence. Although modern approaches overcome some of these problems, most of them require a complex input mask, cannot handle multiple deletion and are unable to remove correlations associated with an object. Recently, a paper has found a new way to combine objects and their effects to create masks containing subjects and effects in a self-supervised manner using only masks and coarse segmentation images. It does this by decomposing a video into a set of RGBA layers representing the appearance of different objects and their effects in the video. Although this requires training one model per video, it can lead to many applications.
1313

1414
**Results:** You can find below my personal results. Details on the training and on the implementation can be found in the following <a href="/assets/pdf/Report_Omnimatte"> pdf </a>. The hardest parts were to pre-process videos by calculating homographies, optical flow, binary masks, etc. and to do a notebook in order to run the code on Colab.
15-
<div class="row">
16-
<div class="col-sm mt-3 mt-md-0">
17-
{% include figure.html path="assets/img/project/DeepL/project_omnimatte_results1.jpg" title="example image" class="img-fluid rounded z-depth-1" %}
18-
</div>
19-
</div>
2015

21-
<div class="row">
22-
<div class="col-sm mt-3 mt-md-0">
23-
{% include figure.html path="assets/img/project/DeepL/project_omnimatte_results1.jpg" title="example image" class="img-fluid rounded z-depth-1" %}
24-
</div>
25-
</div>
26-
<div class="caption">
27-
Our results and the ones of the paper on "Drift chicane".
28-
</div>
29-
30-
<div class="row">
31-
<div class="col-sm mt-3 mt-md-0">
32-
{% include figure.html path="assets/img/project/DeepL/project_omnimatte_results3.jpg" title="example image" class="img-fluid rounded z-depth-1" %}
33-
</div>
34-
</div>
35-
36-
<div class="row">
37-
<div class="col-sm mt-3 mt-md-0">
38-
{% include figure.html path="assets/img/project/DeepL/project_omnimatte_results4.jpg" title="example image" class="img-fluid rounded z-depth-1" %}
39-
</div>
40-
</div>
41-
<div class="caption">
42-
Our results and the ones of the paper on "Blackswan".
16+
<div class="text-center">
17+
<div class="mt-3 mt-md-0">{% include figure.html path="assets/img/project/omnimatte/project_omnimatte_results1.png" title="example image" class="img-fluid rounded z-depth-1" %}</div>
18+
<div class="mt-3 mt-md-0">{% include figure.html path="assets/img/project/omnimatte/project_omnimatte_results1.png" title="example image" class="img-fluid rounded z-depth-1" %}</div>
19+
<div class="caption">Our results and the ones of the paper on "Drift chicane".</div>
20+
21+
<div class="mt-3 mt-md-0">{% include figure.html path="assets/img/project/omnimatte/project_omnimatte_results3.png" title="example image" class="img-fluid rounded z-depth-1" %}</div>
22+
<div class="mt-3 mt-md-0">{% include figure.html path="assets/img/project/omnimatte/project_omnimatte_results4.png" title="example image" class="img-fluid rounded z-depth-1" %}</div>
23+
<div class="caption">Our results and the ones of the paper on "Blackswan".</div>
4324
</div>

_projects/3_project.md

Lines changed: 4 additions & 12 deletions
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22
layout: page
33
title: Occupancy Networks
44
description: 3D reconstruction
5-
img: assets/img/project/RecVis/project_recvis.png
5+
img: assets/img/project/occupancy_net/project_recvis.png
66
importance: 1
77
category: master
88
---
@@ -14,17 +14,9 @@ Several modern methods to reconstruct a 3D meshcan be grouped into two categorie
1414

1515
**Results:** You can find below the results obtained after training DeepSDF, Occupancy network and Shape as Points on the sofa category of ShapeNet. Details on the training and on the implementation can be found in the following <a href="/assets/pdf/Report_DeepSDF.pdf"> pdf. </a>
1616

17-
<div class="row justify-content-md-center">
18-
<div class="col-sm">
19-
</div>
20-
<div class="col-auto-4">
21-
{% include figure.html path="assets/img/project/RecVis/project_recvis_results.jpg" title="example image" class="img-fluid rounded z-depth-1" %}
22-
</div>
23-
<div class="col-sm">
24-
</div>
25-
</div>
26-
<div class="caption">
27-
Groundtruth, DeepSDF, ONET and SAP low noise.
17+
<div class="text-center">
18+
<div class="mt-3 mt-md-0">{% include figure.html path="assets/img/project/occupancy_net/project_recvis_results.png" title="example image" class="img-fluid rounded z-depth-1" %}</div>
19+
<div class="caption">Groundtruth, DeepSDF, ONET and SAP low noise.</div>
2820
</div>
2921

3022
**Notes**:

_projects/4_project.md

Lines changed: 4 additions & 10 deletions
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@
22
layout: page
33
title: KP-conv
44
description: 3D semantic segmentation
5-
img: assets/img/project/IC/profile.png
5+
img: assets/img/project/kpconv/profile.png
66
importance: 1
77
category: master
88
---
@@ -16,15 +16,9 @@ category: master
1616
**Results:** To solve this challenge, I used KP-Conv, a deep neural network build to classify and segment 3D point clouds. I ranked first in the competition with a score of 0.9400 on the private leaderboard. Details on the training and on the implementation can be found in the following <a href="/assets/pdf/Report_IC"> pdf </a>.
1717

1818

19-
<div class="row">
20-
<div class="col-sm mt-3 mt-md-0">
21-
{% include figure.html path="assets/img/project/IC/results_1.png"
22-
title="example image"
23-
class="img-fluid rounded z-depth-1" %}
24-
</div>
25-
</div>
26-
<div class="caption">
27-
Results obtained on the test set. Some classes are over-represented, which corresponds to the distribution of classes in the training scans.
19+
<div class="text-center">
20+
<div class="mt-3 mt-md-0">{% include figure.html path="assets/img/project/kpconv/results_1.png" title="example image" class="img-fluid rounded z-depth-1" %}</div>
21+
<div class="caption">Results obtained on the test set. Some classes are over-represented, which corresponds to the distribution of classes in the training scans.</div>
2822
</div>
2923

3024
**Ressources**: I trained the network on Google Colab pro using a P100 GPU for 10 hours.

_projects/5_project.md

Lines changed: 9 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -6,6 +6,9 @@ img: assets/img/paper/2024_pointbev/pointbev.PNG
66
importance: 1
77
category: paper
88
year: 2024
9+
venue: CVPR
10+
paper_url: https://arxiv.org/abs/2312.00703
11+
code_url: https://github.com/valeoai/PointBeV
912
---
1013

1114
<h1 align="center"> {{page.title}} </h1>
@@ -34,19 +37,19 @@ year: 2024
3437
</p>
3538
</div>
3639

37-
38-
<div class="publication-teaser">
39-
<img src="../../{{ page.image }}" alt="project teaser"/>
40-
</div>
41-
42-
4340
<hr>
4441

4542
<h2 align="center"> Abstract</h2>
4643

4744
<p align="justify">Bird's-eye View (BeV) representations have emerged as the de-facto shared space in driving applications, offering a unified space for sensor data fusion and supporting various downstream tasks. However, conventional models use grids with fixed resolution and range and face computational inefficiencies due to the uniform allocation of resources across all cells. To address this, we propose PointBeV, a novel sparse BeV segmentation model operating on sparse BeV cells instead of dense grids. This approach offers precise control over memory usage, enabling the use of long temporal contexts and accommodating memory-constrained platforms. PointBeV employs an efficient two-pass strategy for training, enabling focused computation on regions of interest. At inference time, it can be used with various memory/performance trade-offs and flexibly adjusts to new specific use cases. PointBeV achieves state-of-the-art results on the nuScenes dataset for vehicle, pedestrian, and lane segmentation, showcasing superior performance in static and temporal settings despite being trained solely with sparse signals. We will release our code along with two new efficient modules used in the architecture: Sparse Feature Pulling, designed for the effective extraction of features from images to BeV, and Submanifold Attention, which enables efficient temporal modeling.</p>
4845

4946
<hr>
47+
48+
<div class="text-center">
49+
<div class="mt-3 mt-md-0">{% include figure.html path="assets/img/paper/2024_pointbev/pointbev.PNG" class="img-fluid rounded z-depth-1" %}</div>
50+
<div class="caption">Architecture.</div>
51+
</div>
52+
5053
<hr>
5154

5255
<h2 align="center">BibTeX</h2>

_projects/6_project.md

Lines changed: 17 additions & 22 deletions
Original file line numberDiff line numberDiff line change
@@ -5,13 +5,14 @@ description: 3D Occupancy
55
img: assets/img/paper/2025_gaussrender/teaser.png
66
importance: 1
77
category: paper
8+
code_url: https://github.com/valeoai/GaussRender
9+
year: 2025
10+
venue: ICCV
11+
paper_url: https://arxiv.org/abs/2502.05040
812
---
913

1014

1115
<h1 align="center"> {{page.title}} </h1>
12-
<!-- Simple call of authors -->
13-
<!-- <h3 align="center"> {{page.authors}} </h3> -->
14-
<!-- Alternatively you can add links to author pages -->
1516
<h3 align="center"> <a href="https://loickch.github.io/">Loïck Chambon</a> &nbsp;&nbsp; <a href="https://eloiz.github.io">Éloi Zablocki</a> &nbsp;&nbsp; <a href="https://boulch.eu/">Alexandre Boulch</a> &nbsp;&nbsp; <a href="https://scholar.google.com/citations?user=QnRpMJAAAAAJ">Mickaël Chen</a> &nbsp;&nbsp; <a href="https://cord.isir.upmc.fr/">Matthieu Cord</a></h3>
1617

1718
<h3 align="center"> {{page.venue}} {{page.year}} </h3>
@@ -38,22 +39,18 @@ category: paper
3839
</div>
3940
</div>
4041

41-
<div class="row">
42-
<div class="col-sm mt-3 mt-md-0" style="padding-right: 5px;padding-left: 5px;">
43-
<img src="assets/img/paper/2025_gaussrender/demo_scene_0003.gif" class="img-fluid rounded z-depth-1"/>
44-
<div class="caption">
45-
Scene visualization 1
42+
<div class="text-center">
43+
<div class="d-flex justify-content-center gap-2">
44+
<div class="flex-fill" style="max-width: 50%; padding: 0 5px;">
45+
<img src="../../assets/img/paper/2025_gaussrender/demo_scene_0003.gif" class="img-fluid rounded z-depth-1"/>
46+
<div class="caption">Scene visualization 1</div>
4647
</div>
47-
</div>
48-
<div class="col-sm mt-3 mt-md-0" style="padding-right: 5px;padding-left: 5px;">
49-
<img src="assets/img/paper/2025_gaussrender/demo_scene_0013.gif" class="img-fluid rounded z-depth-1"/>
50-
<div class="caption">
51-
Scene visualization 2
48+
<div class="flex-fill" style="max-width: 50%; padding: 0 5px;">
49+
<img src="../../assets/img/paper/2025_gaussrender/demo_scene_0013.gif" class="img-fluid rounded z-depth-1"/>
50+
<div class="caption">Scene visualization 2</div>
5251
</div>
5352
</div>
54-
</div>
55-
<div class="caption">
56-
GaussRender is a 3D Occupancy module that can be plugged into any 3D Occupancy model to enhance its predictions and ensure 2D-3D consistency while improving mIoU, IoU, and RayIoU.
53+
<div class="caption mt-3">GaussRender is a 3D Occupancy module that can be plugged into any 3D Occupancy model to enhance its predictions and ensure 2D-3D consistency while improving mIoU, IoU, and RayIoU.</div>
5754
</div>
5855

5956
<hr>
@@ -64,13 +61,11 @@ category: paper
6461

6562
<hr>
6663

67-
<div class="row">
68-
<div class="col-sm mt-3 mt-md-0">
69-
<img src="assets/img/paper/2025_gaussrender/pipeline.png" class="img-fluid rounded z-depth-1"/>
70-
<div class="caption">
71-
GaussRender can be plugged to any model. The core idea is to transform voxels into gaussians before performing a depth and a semantic rendering.
72-
</div>
64+
<div class="text-center">
65+
<div class="mt-3 mt-md-0">
66+
<img src="../../assets/img/paper/2025_gaussrender/pipeline.png" class="img-fluid rounded z-depth-1"/>
7367
</div>
68+
<div class="caption">GaussRender can be plugged to any model. The core idea is to transform voxels into gaussians before performing a depth and a semantic rendering.</div>
7469
</div>
7570

7671
<h2 align="center">Results</h2>

_projects/7_project.md

Lines changed: 10 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -5,6 +5,10 @@ description: Feature Upsampling
55
img: assets/img/paper/2025_jafar/teaser.png
66
importance: 1
77
category: paper
8+
code_url: https://github.com/PaulCouairon/JAFAR
9+
paper_url: https://arxiv.org/pdf/2506.11136
10+
website_url: https://jafar-upsampler.github.io/
11+
year: 2025
812
---
913

1014

@@ -49,7 +53,7 @@ category: paper
4953

5054
<div class="row">
5155
<div class="col-sm mt-3 mt-md-0">
52-
<img src="assets/img/paper/2025_jafar/teaser.png" class="img-fluid rounded z-depth-1"/>
56+
<img src="../../assets/img/paper/2025_jafar/teaser.png" class="img-fluid rounded z-depth-1"/>
5357
<div class="caption">
5458
JAFAR improves metrics on many downstream tasks: semantic segmentation, depth estimation, feature activation, zero-shot open vocabulary, bird's eye view segmentation by upsampling features from any backbone.
5559
</div>
@@ -63,7 +67,7 @@ Once trained, JAFAR can efficiently upsample any backbone features to any resolu
6367

6468
<div class="row">
6569
<div class="col-sm mt-3 mt-md-0">
66-
<img src="assets/img/paper/2025_jafar/pca.png" class="img-fluid rounded z-depth-1"/>
70+
<img src="../../assets/img/paper/2025_jafar/pca.png" class="img-fluid rounded z-depth-1"/>
6771
<div class="caption">
6872
PCA visualization of features from various upsamplers.
6973
</div>
@@ -78,7 +82,7 @@ We perform a linear probing on upsampled features from various upsamplers on man
7882

7983
<div class="row">
8084
<div class="col-sm mt-3 mt-md-0">
81-
<img src="assets/img/paper/2025_jafar/segmentation.png" class="img-fluid rounded z-depth-1"/>
85+
<img src="../../assets/img/paper/2025_jafar/segmentation.png" class="img-fluid rounded z-depth-1"/>
8286
<div class="caption">
8387
Linear probing results for semantic segmentation across various upsamplers.
8488
</div>
@@ -139,7 +143,7 @@ For depth estimation, we evaluate the upsampled features using δ₁ and RMSE me
139143

140144
<div class="row">
141145
<div class="col-sm mt-3 mt-md-0">
142-
<img src="assets/img/paper/2025_jafar/depth.png" class="img-fluid rounded z-depth-1"/>
146+
<img src="../../assets/img/paper/2025_jafar/depth.png" class="img-fluid rounded z-depth-1"/>
143147
<div class="caption">
144148
Linear probing results for depth estimation across various upsamplers.
145149
</div>
@@ -190,7 +194,7 @@ When evaluating Class Activation Maps (CAMs), JAFAR demonstrates improved alignm
190194

191195
<div class="row">
192196
<div class="col-sm mt-3 mt-md-0">
193-
<img src="assets/img/paper/2025_jafar/gradcam.png" class="img-fluid rounded z-depth-1"/>
197+
<img src="../../assets/img/paper/2025_jafar/gradcam.png" class="img-fluid rounded z-depth-1"/>
194198
<div class="caption">
195199
Class Activation Map visualizations across various upsamplers.
196200
</div>
@@ -251,7 +255,7 @@ Even on more complicated tasks and pipelines, JAFAR shows significant improvemen
251255

252256
<div class="row">
253257
<div class="col-sm mt-3 mt-md-0">
254-
<img src="assets/img/paper/2025_jafar/bev.gif" class="img-fluid rounded z-depth-1"/>
258+
<img src="../../assets/img/paper/2025_jafar/bev.gif" class="img-fluid rounded z-depth-1"/>
255259
<div class="caption">
256260
Vehicle segmentation in Bird's Eye View using DINOv2 + JAFAR.
257261
</div>

assets/img/project/Npm3d/project_deephoughvoting.png renamed to assets/img/project/deep_hough_votting/project_deephoughvoting.png

File renamed without changes.

assets/img/project/Npm3d/project_deephoughvoting_results.png renamed to assets/img/project/deep_hough_votting/project_deephoughvoting_results.png

File renamed without changes.

assets/img/project/Npm3d/project_deephoughvoting_results_2.png renamed to assets/img/project/deep_hough_votting/project_deephoughvoting_results_2.png

File renamed without changes.

0 commit comments

Comments
 (0)