Continue training
In the firsrt experiment the model is trained using the SGD optimizer. The value of the optimizer is kept at 0.01 to achieve the baseline evaluation model to carry out further comparison. The experiment is conducted with 30 epochs on the sub –sampled tiles of xView dataset. PR AUC 0.105 at mAP@0.5 | F1 max 0.09 at 0.114 confidence is achieved on the experiment and further tuning is conducted with taking this result as the baseline model. Figure 2 and Figure 3 shows the F1-Confidence curve and the Precision –Recall curve respectively.