Hyun Joon Jung

Jing Shi, Wei Xiong, Zhe Lin et al.

Recent advances in personalized image generation allow a pre-trained text-to-image model to learn a new concept from a set of images. However, existing personalization approaches usually require heavy test-time finetuning for each concept, which is time-consuming and difficult to scale. We propose InstantBooth, a novel approach built upon pre-trained text-to-image models that enables instant text-guided image personalization without any test-time finetuning. We achieve this with several major components. First, we learn the general concept of the input images by converting them to a textual token with a learnable image encoder. Second, to keep the fine details of the identity, we learn rich visual feature representation by introducing a few adapter layers to the pre-trained model. We train our components only on text-image pairs without using paired images of the same concept. Compared to test-time finetuning-based methods like DreamBooth and Textual-Inversion, our model can generate competitive results on unseen concepts concerning language-image alignment, image fidelity, and identity preservation while being 100 times faster.

Hyun Joon Jung, Matthew Lease

We describe methods to predict a crowd worker's accuracy on new tasks based on his accuracy on past tasks. Such prediction provides a foundation for identifying the best workers to route work to in order to maximize accuracy on the new task. Our key insight is to model similarity of past tasks to the target task such that past task accuracies can be optimally integrated to predict target task accuracy. We describe two matrix factorization (MF) approaches from collaborative filtering which not only exploit such task similarity, but are known to be robust to sparse data. Experiments on synthetic and real-world datasets provide feasibility assessment and comparative evaluation of MF approaches vs. two baseline methods. Across a range of data scales and task similarity conditions, we evaluate: 1) prediction error over all workers; and 2) how well each method predicts the best workers to use for each task. Results show the benefit of task routing over random assignment, the strength of probabilistic MF over baseline methods, and the robustness of methods under different conditions.