| 1. |
- Hess, Georg, 1996, et al.
(författare)
-
Masked Autoencoder for Self-Supervised Pre-Training on Lidar Point Clouds
- 2023
-
Ingår i: Proceedings - 2023 IEEE/CVF Winter Conference on Applications of Computer Vision Workshops, WACVW 2023. ; , s. 350-359
-
Konferensbidrag (refereegranskat)abstract
- Masked autoencoding has become a successful pretraining paradigm for Transformer models for text, images, and, recently, point clouds. Raw automotive datasets are suitable candidates for self-supervised pre-training as they generally are cheap to collect compared to annotations for tasks like 3D object detection (OD). However, the development of masked autoencoders for point clouds has focused solely on synthetic and indoor data. Consequently, existing methods have tailored their representations and models toward small and dense point clouds with homogeneous point densities. In this work, we study masked autoencoding for point clouds in an automotive setting, which are sparse and for which the point density can vary drastically among objects in the same scene. To this end, we propose Voxel-MAE, a simple masked autoencoding pre-training scheme designed for voxel representations. We pre-train the backbone of a Transformer-based 3D object detector to reconstruct masked voxels and to distinguish between empty and non-empty voxels. Our method improves the 3D OD performance by 1.75 mAP points and 1.05 NDS on the challenging nuScenes dataset. Further, we show that by pre-training with Voxel-MAE, we require only 40 of the annotated data to outperform a randomly initialized equivalent. Code is available at https://github.com/georghess/voxel-mae.
|
|
| 2. |
- Blasi, Maria, et al.
(författare)
-
Historical and citizen-reported data show shifts in bumblebee phenology over the last century in Sweden
- 2023
-
Ingår i: Biodiversity and Conservation. - : Springer Science and Business Media LLC. - 0960-3115 .- 1572-9710. ; 32:5, s. 1523-1547
-
Tidskriftsartikel (refereegranskat)abstract
- Bumblebees are a key taxon contributing to the provision of crop pollination and ecosystem functioning. However, land use and climate change are two of the main factors causing bee decline across the world. In this study, we investigated how the flight period of bumblebee spring queens has shifted over the last century in Sweden, and to what extent such shifts depended on climate change, landscape context, latitude, and the phenology of bumblebee species. We studied ten species of bumblebees and used observations from museum specimens covering 117 years from the southernmost region in Sweden (Scania), combined with citizen-reported observations during the past 20 years across Sweden. We found that the flight period of bumblebees has advanced by 5 days on average during the last 20 years across Sweden. In the agriculture-dominated region of Scania, we found that in the late 2010s bumblebee spring queen activity in simplified landscapes had advanced by on average 14 days, compared to 100 years ago. In addition, in simplified landscapes the flight period of early species was significantly earlier compared to in complex landscapes. Our results provide knowledge on the intraspecific variation of phenological traits, indicating that early species (often common species) exhibit a higher plastic response to the environment, which may facilitate adaptation to both climate and landscape changes, compared to the late species of which many are declining.
|
|
| 3. |
- Bonagas, Nadilly, et al.
(författare)
-
Pharmacological targeting of MTHFD2 suppresses acute myeloid leukemia by inducing thymidine depletion and replication stress
- 2022
-
Ingår i: NATURE CANCER. - : Springer Science and Business Media LLC. - 2662-1347. ; 3:2, s. 156-
-
Tidskriftsartikel (refereegranskat)abstract
- The folate metabolism enzyme MTHFD2 (methylenetetrahydrofolate dehydrogenase/cyclohydrolase) is consistently overexpressed in cancer but its roles are not fully characterized, and current candidate inhibitors have limited potency for clinical development. In the present study, we demonstrate a role for MTHFD2 in DNA replication and genomic stability in cancer cells, and perform a drug screen to identify potent and selective nanomolar MTHFD2 inhibitors; protein cocrystal structures demonstrated binding to the active site of MTHFD2 and target engagement. MTHFD2 inhibitors reduced replication fork speed and induced replication stress followed by S-phase arrest and apoptosis of acute myeloid leukemia cells in vitro and in vivo, with a therapeutic window spanning four orders of magnitude compared with nontumorigenic cells. Mechanistically, MTHFD2 inhibitors prevented thymidine production leading to misincorporation of uracil into DNA and replication stress. Overall, these results demonstrate a functional link between MTHFD2-dependent cancer metabolism and replication stress that can be exploited therapeutically with this new class of inhibitors. Helleday and colleagues describe a nanomolar MTHFD2 inhibitor that causes replication stress and DNA damage accumulation in cancer cells via thymidine depletion, demonstrating a potential therapeutic strategy in AML tumors in vivo.
|
|
| 4. |
- Hess, Georg, 1996, et al.
(författare)
-
Object Detection as Probabilistic Set Prediction
- 2022
-
Ingår i: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). - Cham : Springer Nature Switzerland. - 1611-3349 .- 0302-9743. - 9783031200793 ; 13670:XVIII, s. 550-566
-
Konferensbidrag (refereegranskat)abstract
- Accurate uncertainty estimates are essential for deploying deep object detectors in safety-critical systems. The development and evaluation of probabilistic object detectors have been hindered by shortcomings in existing performance measures, which tend to involve arbitrary thresholds or limit the detector’s choice of distributions. In this work, we propose to view object detection as a set prediction task where detectors predict the distribution over the set of objects. Using the negative log-likelihood for random finite sets, we present a proper scoring rule for evaluating and training probabilistic object detectors. The proposed method can be applied to existing probabilistic detectors, is free from thresholds, and enables fair comparison between architectures. Three different types of detectors are evaluated on the COCO dataset. Our results indicate that the training of existing detectors is optimized toward non-probabilistic metrics. We hope to encourage the development of new object detectors that can accurately estimate their own uncertainty. Code at https://github.com/georghess/pmb-nll.
|
|
| 5. |
- Meding, Isak, et al.
(författare)
-
You can have your ensemble and run it too - Deep Ensembles Spread Over Time
- 2023
-
Ingår i: Proceedings - 2023 IEEE/CVF International Conference on Computer Vision Workshops, ICCVW 2023. - : IEEE COMPUTER SOC. ; , s. 4022-4031
-
Konferensbidrag (refereegranskat)abstract
- Ensembles of independently trained deep neural networks yield uncertainty estimates that rival Bayesian networks in performance. They also offer sizable improvements in terms of predictive performance over single models. However, deep ensembles are not commonly used in environments with limited computational budget - such as autonomous driving - since the complexity grows linearly with the number of ensemble members. An important observation that can be made for robotics applications, such as autonomous driving, is that data is typically sequential. For instance, when an object is to be recognized, an autonomous vehicle typically observes a sequence of images, rather than a single image. This raises the question, could the deep ensemble be spread over time?In this work, we propose and analyze Deep Ensembles Spread Over Time (DESOT). The idea is to apply only a single ensemble member to each data point in the sequence, and fuse the predictions over a sequence of data points. We implement and experiment with DESOT for traffic sign classification, where sequences of tracked image patches are to be classified. We find that DESOT obtains the benefits of deep ensembles, in terms of predictive and uncertainty estimation performance, while avoiding the added computational cost. Moreover, DESOT is simple to implement and does not require sequences during training. Finally, we find that DESOT, like deep ensembles, outperform single models for out-of-distribution detection.
|
|
