There are eight people in the image. Some of them are facing the camera, namely six, while two have their backs to the camera, and none are sideways. Most people are standing, whereas two are sitting, and none are laying. A few of them are close to the camera, about one, meanwhile the rest are further from the camera. All the individuals appear to have their faces covered, possibly by masks. The most prominent shirt tone is dark, and most of the people are wearing dark-colored pants.
This project was made in the ambit of the final project of the Computer Science degree in Universidade da Beira Interior The objective of this project is to receive images, interpret them, and generate a description using instance segmentation and crowd pose estimation models.
In order to setup this project, two key steps are fundamental, firstly, download the pre-trained weights for both Mask-RCNN and AlphaPose models, The duc_se.pth can be found in this Google Drive, and then place it in alphapose_utils/models/sppe/. The yolov3-spp.weights can be found in this Google Drive, and then place it in alphapose_utils/models/yolo/.
Run train.py to train the Mask-RCNN model.
To properly run this project, add the evaluation dataset to the EvaluationDataset, by adding the images to the EvaluationDataset/seg_input/ folder and, to evaluate the dataset place the ground truth masks in the EvaluationDataset/seg_true_masks/ folder and the ground true interpretations in the EvaluationDataset/prediction_ground_truth/ folder.
To train the segmentation model simply run the train.py script. To fine-tune the training, adjust the parameters.
To run the project run the runProject.sh script or simply run each script individually.