Dubhe/model_measuring/kamal/slim/distillation/vid.py

"""
 Copyright 2020 Tianshu AI Platform. All Rights Reserved.

 Licensed under the Apache License, Version 2.0 (the "License");
 you may not use this file except in compliance with the License.
 You may obtain a copy of the License at

     http://www.apache.org/licenses/LICENSE-2.0

 Unless required by applicable law or agreed to in writing, software
 distributed under the License is distributed on an "AS IS" BASIS,
 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License.
 =============================================================
"""

import numpy as np
import time
import torch.nn as nn
import torch._ops
import torch.nn.functional as F
from .kd import KDDistiller
from kamal.utils import set_mode
from kamal.core.tasks.loss import KDLoss

class VIDDistiller(KDDistiller):
    def __init__(self, logger=None, tb_writer=None ):
        super(VIDDistiller, self).__init__( logger, tb_writer )

    def setup(self, student, teacher, dataloader, optimizer, regressor_l, T=1.0, alpha=1.0, beta=1.0, gamma=1.0, stu_hooks=[], tea_hooks=[], out_flags=[], device=None):
        super( VIDDistiller, self ).setup( 
            student, teacher, dataloader, optimizer, T=T, alpha=alpha, beta=beta, gamma=gamma, device=device )
        self.regressor_l = regressor_l
        self.stu_hooks = stu_hooks
        self.tea_hooks = tea_hooks
        self.out_flags = out_flags
        self.regressor_l = [regressor.to(self.device).train() for regressor in self.regressor_l]

    def additional_kd_loss(self, engine, batch):
        feat_s =  [f.feat_out if flag else f.feat_in for (f, flag) in zip(self.stu_hooks, self.out_flags)]
        feat_t = [f.feat_out.detach() if flag else f.feat_in for (f, flag) in zip(self.tea_hooks, self.out_flags)]
        g_s = feat_s[1:-1]
        g_t = feat_t[1:-1]
        return sum([c(f_s, f_t) for f_s, f_t, c in zip(g_s, g_t, self.regressor_l)])

class VIDRegressor(nn.Module):
    def __init__(self,
                num_input_channels,
                num_mid_channel,
                num_target_channels,
                init_pred_var=5.0,
                eps=1e-5):
        super(VIDRegressor, self).__init__()

        def conv1x1(in_channels, out_channels, stride=1):
            return nn.Conv2d(
                in_channels, out_channels,
                kernel_size=1, padding=0,
                bias=False, stride=stride)

        self.regressor = nn.Sequential(
            conv1x1(num_input_channels, num_mid_channel),
            nn.ReLU(),
            conv1x1(num_mid_channel, num_mid_channel),
            nn.ReLU(),
            conv1x1(num_mid_channel, num_target_channels),
        )
        self.log_scale = torch.nn.Parameter(
            np.log(np.exp(init_pred_var-eps)-1.0) * torch.ones(num_target_channels)
            )
        self.eps = eps

    def forward(self, input, target):
        # pool for dimentsion match
        s_H, t_H = input.shape[2], target.shape[2]
        if s_H > t_H:
            input = F.adaptive_avg_pool2d(input, (t_H, t_H))
        elif s_H < t_H:
            target = F.adaptive_avg_pool2d(target, (s_H, s_H))
        else:
            pass
        pred_mean = self.regressor(input)
        pred_var = torch.log(1.0+torch.exp(self.log_scale))+self.eps
        pred_var = pred_var.view(1, -1, 1, 1)
        neg_log_prob = 0.5*(
            (pred_mean-target)**2/pred_var+torch.log(pred_var)
            )
        loss = torch.mean(neg_log_prob)
        return loss