Imlab.postech.ac.kr
Project 5: Object Detection (Due: Jun. 25)The goal of this project is to train and test famous object detectors (Faster RCNN, Mask RCNN and YOLOv3) on Pascal VOC 2007 dataset. This project requires many computations, so GPUs are essential. In addition, to facilitate your tasks, we recommend you to use Ubuntu 16.04 because there are effective shell scripts for using networks.Installation GuidesInstall Ubuntu 16.04 desktop image (ref: )Install Python version 3.6 (ref: )Install Pytorch 1.1 (or higher) on python 3.6 (ref: )Download Pascal VOC 2007 dataset (ref: )Problem #1: Training and Testing Faster RCNN with ResNet-50 on Pascal VOC 2007 DatasetBasic guideInstall “mmdetection” (ref: )In this repository, README.md provides a detailed descriptionsIf you have any problems, refer ‘Issues’ in this repositoryTraining and testing Faster RCNNUnderstand the concepts of Faster RCNN (ref: )Implement Faster RCNN based on “mmdetection”Train and test Faster RCNN with ResNet-50 on Pascal VOC 2007 dataset Report experimental resultsFill out the following blanks in terms of mean average precisions (mAP) and inference times (FPS) where mAP@# means that a prediction is positive if IoU ≥ # and discuss your experimental results.?mAP@0.5mAP@0.6mAP@0.7mAP@0.8mAP@0.9FPSFaster RCNN??????Try the efforts to improve the performances on your network models, such as your learning techniques or your network improvements that are not provided by basic codesShow learning curves for training and validationShow your source codes and trained model parametersProblem #2: Training and Testing Mask RCNN with ResNet-50 on Pascal VOC 2007 Dataset1) Basic guide1. Install “mmdetection” (ref: )In this repository, README.md provides a detailed descriptionsIf you have any problems, refer ‘Issues’ in this repository2) Training and testing Mask RCNNUnderstand the concepts of Mask RCNN (ref: )Implement Mask RCNN based on “mmdetection”Train and test Mask RCNN with ResNet-50 on Pascal VOC 2007 dataset 3) Report experimental resultsFill out the following blanks in terms of mean average precisions (mAP) and inference times (FPS) where mAP@# means that a prediction is positive if IoU ≥ # and discuss your experimental results.?mAP@0.5mAP@0.6mAP@0.7mAP@0.8mAP@0.9FPSMask RCNN??????Try the efforts to improve the performances on your network models, such as your learning techniques or your network improvements that are not provided by basic codesShow learning curves for training and validationShow your source codes and trained model parametersProblem #3: Training and Testing YOLOv3 with darknet53 on Pascal VOC 2007 datasetBasic guideInstall “PyTorch-YOLOv3” (ref: )In this repository, README.md provides a detailed descriptionsIf you have any problems, refer ‘Issues’ in this repositoryTraining and testing YOLOv3Understand the concepts of YOLOv3 (ref: )Implement YOLOv3 based on “PyTorch-YOLOv3”Train and test YOLOv3 with darknet53 on Pascal VOC 2007 dataset Download the pre-trained darknet53 on ImageNet 1k (ref: )3) Report experimental resultsFill out the following blanks in terms of mean average precisions (mAP) and inference times (FPS) where mAP@# means that a prediction is positive if IoU ≥ # and discuss your experimental results.?mAP@0.5mAP@0.6mAP@0.7mAP@0.8mAP@0.9FPSYoLov3??????Try the efforts to improve the performances on your network models, such as your learning techniques or your network improvements that are not provided by basic codesShow learning curves for training and validationShow your source codes and trained model parameters ................
................
In order to avoid copyright disputes, this page is only a partial summary.
To fulfill the demand for quickly locating and searching documents.
It is intelligent file search solution for home and business.