Object detection tasks often involve challenges such as incorrect labeling and managing multiple bounding box predictions, which complicate the evaluation process. To address these challenges, the article introduces a streamlined system using Comet and Aquarium, which allows for efficient model evaluation without extensive coding. Comet, an MLOps platform, and Aquarium, an ML data management platform, facilitate the tracking, exploration, and improvement of datasets by enabling users to log data, track and version datasets, and analyze model predictions. By using tools like the Comet Artifacts and Aquarium's embedding viewer and confusion matrix, users can identify labeling errors and problematic data points. This approach enhances the evaluation process by allowing for ad-hoc metric computation and dataset updates through Webhooks, reducing the need for manual intervention and making the process quicker and more standardized. The article demonstrates this system with a practical example using the DOTA dataset and a FasterRCNN model, emphasizing the benefits of interactive data exploration and error correction.