Face, a non-intrusive recognition modality, is an ideal candidate for identifying criminals or performing general-purpose person identification. On top of that, faces are not only related to identity but other essential attributes such as age and gender can also be extracted. Due to such potential, face recognition has received tremendous attention, yet face recognition from a distance remains challenging. To empower the face recognition research, we have collected a novel unconstrained video face dataset namely MobileFaces across various distances using mobile phones. Utilizing the proposed dataset, we have performed extensive experiments on face recognition, including verification and identification using state-of-the-art (SOTA) deep face recognition networks. Further, we have evaluated the robustness of current SOTA deep face attributes prediction networks to demonstrate whether the challenge lies in face recognition only or whether the existing algorithms are vulnerable in predicting facial attributes such as age and gender. The results suggest that the existing algorithms are ineffective not only in identifying the identity of the subjects but also fail to detect face attributes when the images are captured in unconstrained environments. For example, deep face networks yield the best macro average accuracy of \(\textbf{65}\) % and an F-1 score of \(\mathbf {0.48}\) when asked to predict gender on the collected dataset at a distance of 10 m. Based on the comparison with existing unconstrained face datasets and analysis of the effectiveness of image super-resolution techniques, it is showcased that the proposed dataset is significantly challenging compared to them, and hence, we believe that the presence of our dataset can advance the development of unconstrained face recognition algorithms.