Object-aware Contrastive Learning for Debiased Scene Representation

Contrastive self-supervised learning has shown impressive results in learning visual representations from unlabeled images by enforcing invariance against different data augmentations. However, the learned representations are often contextually biased to the spurious scene correlations of different objects or object and background, which may harm their generalization on the downstream tasks. To tackle the issue, we develop a novel object-aware contrastive learning framework that first (a) localizes objects in a self-supervised manner and then (b) debias scene correlations via appropriate data augmentations considering the inferred object locations. For (a), we propose the contrastive class activation map (ContraCAM), which finds the most discriminative regions (e.g., objects) in the image compared to the other images using the contrastively trained models. We further improve the ContraCAM to detect multiple objects and entire shapes via an iterative refinement procedure. For (b), we introduce two data augmentations based on ContraCAM, object-aware random crop and background mixup, which reduce contextual and background biases during contrastive self-supervised learning, respectively. Our experiments demonstrate the effectiveness of our representation learning framework, particularly when trained under multi-object images or evaluated under the background (and distribution) shifted images.

Simultaneous observation of water and class I methanol masers toward class II methanol maser sources

We present a simultaneous single-dish survey of 22 GHz water maser and 44 GHz and 95 GHz class I methanol masers toward 77 6.7 GHz class II methanol maser sources, which were selected from the Arecibo methanol maser Galactic plane survey (AMGPS) catalog.Water maser emission is detected in 39 (51%) sources, of which 15 are new detections. Methanol maser emission at 44 GHz and 95 GHz is found in 25 (32%) and 19 (25%) sources, of which 21 and 13 sources are newly detected, respectively. We find 4 high-velocity (> 30 km/s) water maser sources, including 3 dominant blue- or redshifted outflows.The 95 GHz masers always appear with the 44 GHz maser emission. They are strongly correlated with 44 GHz masers in velocity, flux density, and luminosity, while they are not correlated with either water or 6.7 GHz class II methanol masers. The average peak flux density ratio of 95 GHz to 44 GHz masers is close to unity, which is two times higher than previous estimates. The flux densities of class I methanol masers are more closely correlated with the associated BGPS core mass than those of water or class II methanol masers. Using the large velocity gradient (LVG) model and assuming unsaturated class I methanol maser emission, we derive the fractional abundance of methanol to be in a range of 4.2*10^-8 to 2.3*10^-6, with a median value of 3.3pm2.7*10^-7.