import unittest import torch from evaluate.coco_eval import run_eval from lib.network.rtpose_vgg import get_model, use_vgg from torch import load #Notice, if you using the with torch.autograd.no_grad(): # this path is with respect to the root of the project weight_name = '/home/tensorboy/Downloads/pose_model.pth' state_dict = torch.load(weight_name) model = get_model(trunk='vgg19') #model = torch.nn.DataParallel(model).cuda() model.load_state_dict(state_dict) model.eval() model.float() model = model.cuda() # The choice of image preprocessing include: 'rtpose', 'inception', 'vgg' and 'ssd'. # If you use the converted model from caffe, it is 'rtpose' preprocess, the model trained in # this repo used 'vgg' preprocess run_eval(image_dir= '/home/tensorboy/data/coco/images/val2017', anno_file = '/home/tensorboy/data/coco/annotations/person_keypoints_val2017.json', vis_dir = '/home/tensorboy/data/coco/images/vis_val2017', model=model, preprocess='rtpose')
import unittest import torch from evaluate.coco_eval import run_eval from network.rtpose_vgg import get_model, use_vgg from torch import load #Notice, if you using the with torch.autograd.no_grad(): # this path is with respect to the root of the project weight_name = './network/weight/best_pose.pth' state_dict = torch.load(weight_name) model = get_model(trunk='vgg19') model = torch.nn.DataParallel(model).cuda() model.load_state_dict(state_dict) model.eval() model.float() model = model.cuda() # The choice of image preprocessing include: 'rtpose', 'inception', 'vgg' and 'ssd'. # If you use the converted model from caffe, it is 'rtpose' preprocess, the model trained in # this repo used 'vgg' preprocess run_eval(image_dir='/data/coco/images/', anno_dir='/data/coco', vis_dir='/data/coco/vis', image_list_txt='./evaluate/image_info_val2014_1k.txt', model=model, preprocess='vgg')
import unittest import torch from evaluate.coco_eval import run_eval from network.rtpose_vgg import get_model, use_vgg from torch import load #Notice, if you using the with torch.autograd.no_grad(): # this path is with respect to the root of the project # weight_name = './network/weight/best_pose.pth' weight_name = '/home/jie/kiktech-seg/OpenPoseV1/network/weight/pose_model.pth' state_dict = torch.load(weight_name) model = get_model(trunk='vgg19') model.load_state_dict(state_dict) model = torch.nn.DataParallel(model).cuda() model.eval() model.float() model = model.cuda() # The choice of image preprocessing include: 'rtpose', 'inception', 'vgg' and 'ssd'. # If you use the converted model from caffe, it is 'rtpose' preprocess, the model trained in # this repo used 'vgg' preprocess run_eval(image_dir='/home/jie/dataset/COCO/images/', anno_dir='/home/jie/dataset/COCO/', vis_dir='./coco_vis', image_list_txt='image_info_val2014_1k.txt', model=model, preprocess='vgg')
from lib.network.openpose import OpenPose_Model, use_vgg from torch import load #Notice, if you using the with torch.autograd.no_grad(): # this path is with respect to the root of the project weight_name = '/data/rtpose/rtpose_lr001/1/_ckpt_epoch_82.ckpt' state_dict = torch.load(weight_name)['state_dict'] new_state_dict = OrderedDict() for k, v in state_dict.items(): name = k[6:] new_state_dict[name] = v model = get_model(trunk='vgg19') #model = openpose = OpenPose_Model(l2_stages=4, l1_stages=2, paf_out_channels=38, heat_out_channels=19) #model = torch.nn.DataParallel(model).cuda() model.load_state_dict(new_state_dict) model.eval() model.float() model = model.cuda() # The choice of image preprocessing include: 'rtpose', 'inception', 'vgg' and 'ssd'. # If you use the converted model from caffe, it is 'rtpose' preprocess, the model trained in # this repo used 'vgg' preprocess run_eval(image_dir='/data/coco/images/val2017', anno_file='/data/coco/annotations/person_keypoints_val2017.json', vis_dir='/data/coco/images/vis_val2017', model=model, preprocess='vgg')