This paper addresses the efficacy of conducting early fundus screenings to mitigate the risk of vision loss from ophthalmic diseases and presents a solution utilizing deep learning for the ocular disease’s diagnosis, which often involve multiple diseases in a single eye. The research utilizes the RFMiD dataset, focusing on 25 out of its 46 classes while grouping the less frequent classes into a single category termed "others." Preprocessing steps, including background cropping and histogram equalization are employed to improve the focus on the fundus and improve feature clarity. The multi-label classification model is based on the InceptionV3 architecture and is rigorously validated on 20% of the dataset. Performance evaluation employs five key measurements: accuracy, precision, recall, F1-score, and area under the curve resulting in impressive outcomes of 97.04%, 86.62%, 68.04%, 76.21%, and 0.936 respectively. These findings highlight how well the suggested system works for diagnosing eye diseases.