Search for: All records

Object detectors often perform poorly on data that differs from their training set. Domain adaptive object detection (DAOD) methods have recently demonstrated strong results on addressing this challenge. Unfortunately, we identify systemic benchmarking pitfalls that call past results into question and hamper further progress: (a) Overestimation of performance due to underpowered baselines, (b) Inconsistent implementation practices preventing transparent comparisons of methods, and (c) Lack of generality due to outdated backbones and lack of diversity in benchmarks. We address these problems by introducing: (1) A unified benchmarking and implementation framework, Align and Distill (ALDI), enabling comparison of DAOD methods and supporting future development, (2) A fair and modern training and evaluation protocol for DAOD that addresses benchmarking pitfalls, (3) A new DAOD benchmark dataset, CFC-DAOD, increasing the diversity of available DAOD benchmarks, and (4) A new method, ALDI++, that achieves state-of-the-art results by a large margin. ALDI++ outperforms the previous state-of-the-art by +3.5 AP50 on Cityscapes Foggy Cityscapes, +5.7 AP50 on Sim10k Cityscapes (where ours is the only method to outperform a fair baseline), and +0.6 AP50 on CFC-DAOD. ALDI and ALDI++ are architecture-agnostic, setting a new state-of-the-art for YOLO and DETR-based DAOD as well without additional hyperparameter tuning. Our framework, dataset, and method offer a critical reset for DAOD and provide a strong foundation for future research. 

Specialized domain knowledge is often necessary to ac- curately annotate training sets for in-depth analysis, but can be burdensome and time-consuming to acquire from do- main experts. This issue arises prominently in automated behavior analysis, in which agent movements or actions of interest are detected from video tracking data. To reduce annotation effort, we present TREBA: a method to learn annotation-sample efficient trajectory embedding for be- havior analysis, based on multi-task self-supervised learn- ing. The tasks in our method can be efficiently engineered by domain experts through a process we call “task program- ming”, which uses programs to explicitly encode structured knowledge from domain experts. Total domain expert effort can be reduced by exchanging data annotation time for the construction of a small number of programmed tasks. We evaluate this trade-off using data from behavioral neuro- science, in which specialized domain knowledge is used to identify behaviors. We present experimental results in three datasets across two domains: mice and fruit flies. Using embeddings from TREBA, we reduce annotation burden by up to a factor of 10 without compromising accuracy com- pared to state-of-the-art features. Our results thus suggest that task programming and self-supervision can be an ef- fective way to reduce annotation effort for domain experts. 

We propose a method for learning the posture and struc- ture of agents from unlabelled behavioral videos. Start- ing from the observation that behaving agents are gener- ally the main sources of movement in behavioral videos, our method, Behavioral Keypoint Discovery (B-KinD), uses an encoder-decoder architecture with a geometric bottle- neck to reconstruct the spatiotemporal difference between video frames. By focusing only on regions of movement, our approach works directly on input videos without requir- ing manual annotations. Experiments on a variety of agent types (mouse, fly, human, jellyfish, and trees) demonstrate the generality of our approach and reveal that our dis- covered keypoints represent semantically meaningful body parts, which achieve state-of-the-art performance on key- point regression among self-supervised methods. Addition- ally, B-KinD achieve comparable performance to supervised keypoints on downstream tasks, such as behavior clas- sification, suggesting that our method can dramatically re- duce model training costs vis-a-vis supervised methods.