Learning to Overcome Noise in Weak Caption Supervision for Object Detection.

We propose the first mechanism to train object detection models from weak supervision in the form of captions at the image level. Language-based supervision for detection is appealing and inexpensive: many blogs with images and descriptive text written by human users exist. However, there is significant noise in this supervision: captions do not mention all objects that are shown, and may mention extraneous concepts. We first propose a technique to determine which image-caption pairs provide suitable signal for supervision. We further propose several complementary mechanisms to extract image-level pseudo labels for training from the caption. Finally, we train an iterative weakly-supervised object detection model from these image-level pseudo labels. We use captions from four datasets (COCO, Flickr30K, MIRFlickr1M, and Conceptual Captions) whose level of noise varies. We evaluate our approach on two object detection datasets. Weighting the labels extracted from different captions provides a boost over treating all captions equally. Further, our primary proposed technique for inferring pseudo labels for training at the image level, outperforms alternative techniques under a wide variety of settings. Both techniques generalize to datasets beyond the one they were trained on.

Complex wavelets for extended depth-of-field: a new method for the fusion of multichannel microscopy images.

Microscopy imaging often suffers from limited depth-of-field. However, the specimen can be "optically sectioned" by moving the object along the optical axis. Then different areas appear in focus in different images. Extended depth-of-field is a fusion algorithm that combines those images into one single sharp composite. One promising method is based on the wavelet transform. Here, we show how the wavelet-based image fusion technique can be improved and easily extended to multichannel data. First, we propose the use of complex-valued wavelet bases, which seem to outperform traditional real-valued wavelet transforms. Second, we introduce a way to apply this technique for multichannel images that suppresses artifacts and does not introduce false colors, an important requirement for multichannel optical microscopy imaging. We evaluate our method on simulated image stacks and give results relevant to biological imaging.