Background: Laparoscopic cholecystectomy is the standard of care for symptomatic cholelithiasis and cholecystitis and is one of the most performed general surgery procedures.. The potential applications of AI in surgery are broad and continuously progressing, including phase recognition, instrument tracking, anatomical recognition, intraoperative recognition of pathology. A major obstacle to the progress of AI applications in surgery is the development of large training datasets taking into consideration variability in pathological states and anatomy. Furthermore, existing datasets lack the granularity to detect all anatomical structures necessary for the development of augmented reality environments and recognition of impending complication or injury. Methods: A dataset of 4,868 frames taken from 50 laparoscopic cholecystectomy operations were comprehensively labelled using a protocol of 27 classes. DeepLabV3+ neural network was trained using the dataset and tested on 20% holdout frames. Results: The network achieved a mean accuracy metric of 0.56 and a mean F1 score of 0.85. Liver and gallbladder were the best recognised anatomical structure, achieving an F1 of 0.91 and 0.87 while Rouviere’s sulcus was poorly recognised with an F1 score of 0.33 Conclusion: Building from our pilot experiments, we outline the development of a highly detailed labelled laparoscopic cholecystectomy dataset plus novel data processing techniques for the training of AI algorithms in laparoscopic cholecystectomy. This work will underpin the development of technology aimed guiding surgeons, providing intraoperative feedback, and serving as an adjunct to surgical training.