def test(
dataloader,
model,
save_dir="",
device="cuda",
distributed=False,
):
if distributed:
model = model.module
dataset = dataloader.dataset
device = torch.device(device)
num_devices = get_world_size()
detections = compute_on_dataset(model, dataloader, device)
synchronize()
predictions = _accumulate_predictions_from_multiple_gpus(detections)
if not is_main_process():
return
return dataset.evaluation(predictions, str(save_dir))
def test(
dataloader, model, save_dir="", device="cuda", distributed=False,
):
if distributed:
model = model.module
dataset = dataloader.dataset
device = torch.device(device)
num_devices = get_world_size()
# detections = compute_on_dataset(model, dataloader, device)
# torch.save(detections, "/home/jty/pcdet/final_predictions_debug.pkl")
detections=torch.load( "/home/jty/pcdet/final_predictions_debug.pkl")
import pickle
# with open(
# '/home/jty/pcdet/pvrcnn/OpenLidarPerceptron/output/cfgs/kitti_models/pvrcnn_deecamp/default/eval/epoch_24/val/default/result.pkl',
# 'rb') as input_file:
# dt = pickle.load(input_file)
with open(
'/home/jty/Downloads/result.pkl',
'rb') as input_file:
dt = pickle.load(input_file)
synchronize()
# predictions = _accumulate_predictions_from_multiple_gpus(dt)
# predictions={}
if not is_main_process():
return
return dataset.evaluation(dt, str(save_dir))
def _accumulate_predictions_from_multiple_gpus(predictions_per_gpu):
all_predictions = all_gather(predictions_per_gpu)
if not is_main_process():
return
predictions = {}
for p in all_predictions:
predictions.update(p)
return predictions
