# Copyright 2022-present, Lorenzo Bonicelli, Pietro Buzzega, Matteo Boschini, Angelo Porrello, Simone Calderara.
# All rights reserved.
# This source code is licensed under the license found in the
# LICENSE file in the root directory of this source tree.
import torch
from models.utils.continual_model import ContinualModel
from utils.args import add_rehearsal_args, ArgumentParser
from utils.buffer import Buffer
from models.star_utils.star_perturber import Perturber, add_perturb_args
[docs]
class ErACESTAR(ContinualModel):
NAME = 'er_ace_star'
COMPATIBILITY = ['class-il', 'task-il']
[docs]
@staticmethod
def get_parser(parser) -> ArgumentParser:
# add arguments for STAR
add_perturb_args(parser)
add_rehearsal_args(parser)
return parser
def __init__(self, backbone, loss, args, transform, dataset=None):
super().__init__(backbone, loss, args, transform, dataset)
self.buffer = Buffer(self.args.buffer_size)
self.pert = Perturber(self)
self.seen_so_far = torch.tensor([]).long().to(self.device)
[docs]
def observe(self, inputs, labels, not_aug_inputs, epoch=None):
present = labels.unique()
self.seen_so_far = torch.cat([self.seen_so_far, present]).unique()
self.opt.zero_grad()
if not self.buffer.is_empty():
# STAR here
buf_inputs, buf_labels = self.buffer.get_data(
self.args.minibatch_size, transform=self.transform)
self.pert(buf_inputs, buf_labels)
# normal er_ace resumes
logits = self.net(inputs)
mask = torch.zeros_like(logits)
mask[:, present] = 1
if self.seen_so_far.max() < (self.num_classes - 1):
mask[:, self.seen_so_far.max():] = 1
if self.current_task > 0:
logits = logits.masked_fill(mask == 0, torch.finfo(logits.dtype).min)
loss = self.loss(logits, labels)
loss_re = torch.tensor(0.)
if self.current_task > 0:
# sample from buffer
buf_inputs, buf_labels = self.buffer.get_data(
self.args.minibatch_size, transform=self.transform, device=self.device)
loss_re = self.loss(self.net(buf_inputs), buf_labels)
loss += loss_re
loss.backward()
self.opt.step()
self.buffer.add_data(examples=not_aug_inputs,
labels=labels)
return loss.item()