Bark-101 is a challenging dataset made of 101 bark classes. It has been designed on top of the LifeCLEF challenge to evaluate both texture and bark recognition algorithms in the wild. Bark-101 answers the lack of extended and segmented dataset in the context of bark recognition.