trainer.py

from collections import namedtuple
import time
from torch.nn import functional as F
from model.utils.creator_tool import AnchorTargetCreator, ProposalTargetCreator

from torch import nn
import torch as t
from torch.autograd import Variable
from utils import array_tool as at
from utils.vis_tool import Visualizer

from utils.config import opt
from torchnet.meter import ConfusionMeter, AverageValueMeter


import numpy as np




# create a namedtuple
LossTuple = namedtuple('LossTuple',
                       ['rpn_loc_loss',
                        'rpn_cls_loss',
                        'roi_loc_loss',
                        'roi_cls_loss',
                        'total_loss'
                        ])


class FasterRCNNTrainer(nn.Module):
    """wrapper for conveniently training. return losses

    The losses include:

    * :obj:`rpn_loc_loss`: The localization loss for Region Proposal Network (RPN).
    * :obj:`rpn_cls_loss`: The classification loss for RPN.
    * :obj:`roi_loc_loss`: The localization loss for the head module.
    * :obj:`roi_cls_loss`: The classification loss for the head module.
    * :obj:`total_loss`: The sum of 4 loss above.

    Args:
        faster_rcnn (model.FasterRCNN):
            A Faster R-CNN model that is going to be trained.
    """

    def __init__(self, faster_rcnn):
        super(FasterRCNNTrainer, self).__init__()

        self.faster_rcnn = faster_rcnn
        self.rpn_sigma = opt.rpn_sigma
        self.roi_sigma = opt.roi_sigma

        # target creator create gt_bbox gt_label etc as training targets. 
        self.anchor_target_creator = AnchorTargetCreator()
        self.proposal_target_creator = ProposalTargetCreator()

        self.optimizer = self.faster_rcnn.get_optimizer()
        # visdom wrapper
        self.vis = Visualizer(env=opt.env)

        # indicators for training status
        self.rpn_cm = ConfusionMeter(2)
        self.roi_cm = ConfusionMeter(21)
        self.meters = {k: AverageValueMeter() for k in LossTuple._fields}  # average loss

    def forward(self, imgs, bboxes, labels, scale):
        """Forward Faster R-CNN and calculate losses.

        Here are notations used.

        * :math:`N` is the batch size.
        * :math:`R` is the number of bounding boxes per image.

        Currently, only :math:`N=1` is supported.

        Args:
            imgs (~torch.autograd.Variable): A variable with a batch of images.
            bboxes (~torch.autograd.Variable): A batch of bounding boxes.
                Its shape is :math:`(N, R, 4)`.
            labels (~torch.autograd..Variable): A batch of labels.
                Its shape is :math:`(N, R)`. The background is excluded from
                the definition, which means that the range of the value
                is :math:`[0, L - 1]`. :math:`L` is the number of foreground
                classes.
            scale (float): Amount of scaling applied to
                the raw image during preprocessing.

        Returns:
            namedtuple of 5 losses
        """
        n = bboxes.shape[0] # number of input images one time
        if n != 1:
            raise ValueError('Currently only batch size 1 is supported.')

        _, _, H, W = imgs.shape # should be (1,3,H,W)
        img_size = (H, W)

        # need more feature maps here when you are trying to use features of different scale
        features = self.faster_rcnn.extractor(imgs)

        rpn_locs, rpn_scores, rois, search_regions, roi_indices, anchor = self.faster_rcnn.rpn(features, img_size, scale)

        # Since batch size is one, convert variables to singular form
        # different parameters here :
        #     num_boxes : number of ground truth bounding boxes in a image.
        #     num_anchors : number of anchors in images(or to say in a feature map).
        #     num_rois : number of ROIs that are generated by RPN, which will be used in Fast RCNN.
        bbox = bboxes[0]                # shape (num_boxes, 4)
        label = labels[0]               # shape (num_boxes,)
        rpn_score = rpn_scores[0]       # shape (num_anchors,)
        rpn_loc = rpn_locs[0]           # shape (num_anchors, 4)
        roi = rois                      # shape (num_rois, 4)
        search_region = search_regions  # shape (num_rois, 4)

        # Sample RoIs and forward
        # it's fine to break the computation graph of rois, 
        # consider them as constant input
        sample_roi, sample_search_region, (Tx,Ty), gt_roi_label = self.proposal_target_creator(roi,
                                                                                              search_region,
                                                                                              at.tonumpy(bbox),
                                                                                              at.tonumpy(label))

        # NOTE it's all zero because now it only support for batch=1 now
        sample_roi_index = t.zeros(len(sample_roi))
        (px, py), roi_score = self.faster_rcnn.head(features,
                                                    sample_roi,
                                                    sample_search_region,
                                                    sample_roi_index)


        # ------------------ RPN losses -------------------#
        gt_rpn_loc, gt_rpn_label = self.anchor_target_creator(at.tonumpy(bbox),
                                                              anchor,
                                                              img_size)
        gt_rpn_label = at.tovariable(gt_rpn_label).long()
        gt_rpn_loc = at.tovariable(gt_rpn_loc)
        rpn_loc_loss = _fast_rcnn_loc_loss(rpn_loc,
                                           gt_rpn_loc,
                                           gt_rpn_label.data,
                                           self.rpn_sigma)
        
        # NOTE: default value of ignore_index is -100 ...
        rpn_cls_loss = F.cross_entropy(rpn_score, gt_rpn_label.cuda(), ignore_index=-1)
        _gt_rpn_label = gt_rpn_label[gt_rpn_label > -1]
        _rpn_score = at.tonumpy(rpn_score)[at.tonumpy(gt_rpn_label) > -1]
        self.rpn_cm.add(at.totensor(_rpn_score, False), _gt_rpn_label.data.long())


        # ------------------ ROI losses (fast rcnn loss) -------------------#
        n_sample = px.shape[0]
        # (px, py) and (Tx, Ty) are to be used to caculate loss ：roi_loc_loss

        Tx = at.tovariable(Tx).float()
        Ty = at.tovariable(Ty).float()

        print("px is ", px)
        # print("max of px is ", t.max(px))
        # print("min of px is ", t.min(px))
        # print(t.max(Tx))
        # print(t.max(Ty))
        # print(Tx.shape, Ty.shape, px.shape, py.shape)

        roi_loc_loss = _LocNet_loss(Tx, Ty, px, py, gt_roi_label.data, self.roi_sigma)

        
        gt_roi_label = at.tovariable(gt_roi_label).long()
        roi_cls_loss = nn.CrossEntropyLoss()(roi_score, gt_roi_label.cuda())

        self.roi_cm.add(at.totensor(roi_score, False), gt_roi_label.data.long())

        
        losses = [rpn_loc_loss, rpn_cls_loss, roi_loc_loss, roi_cls_loss]

        print("losses", losses)

        losses = losses + [sum(losses)]

        return LossTuple(*losses)  # return a namedtuple

    def train_step(self, imgs, bboxes, labels, scale):
        self.optimizer.zero_grad()
        losses = self.forward(imgs, bboxes, labels, scale) # losses is a namedtuple
        losses.total_loss.backward()                       # use total_loss to backprop
        self.optimizer.step()
        self.update_meters(losses)
        return losses

    def save(self, save_optimizer=False, save_path=None, **kwargs):
        """serialize models include optimizer and other info
        return path where the model-file is stored.

        Args:
            save_optimizer (bool): whether save optimizer.state_dict().
            save_path (string): where to save model, if it's None, save_path
                is generate using time str and info from kwargs.
        
        Returns:
            save_path(str): the path to save models.
        """
        save_dict = dict()

        save_dict['model'] = self.faster_rcnn.state_dict()
        save_dict['config'] = opt._state_dict()
        save_dict['other_info'] = kwargs
        save_dict['vis_info'] = self.vis.state_dict()

        if save_optimizer:
            save_dict['optimizer'] = self.optimizer.state_dict()

        if save_path is None:
            timestr = time.strftime('%m%d%H%M')
            save_path = 'checkpoints/fasterrcnn_%s' % timestr
            for k_, v_ in kwargs.items():
                save_path += '_%s' % v_

        t.save(save_dict, save_path)
        self.vis.save([self.vis.env])
        return save_path

    def load(self, path, load_optimizer=True, parse_opt=False, ):
        state_dict = t.load(path)
        if 'model' in state_dict:
            self.faster_rcnn.load_state_dict(state_dict['model'])
        else:  # legacy way, for backward compatibility
            self.faster_rcnn.load_state_dict(state_dict)
            return self
        if parse_opt:
            opt._parse(state_dict['config'])
        if 'optimizer' in state_dict and load_optimizer:
            self.optimizer.load_state_dict(state_dict['optimizer'])
        return self

    def update_meters(self, losses):
        loss_d = {k: at.scalar(v) for k, v in losses._asdict().items()}
        for key, meter in self.meters.items():
            meter.add(loss_d[key])

    def reset_meters(self):
        for key, meter in self.meters.items():
            meter.reset()
        self.roi_cm.reset()
        self.rpn_cm.reset()

    def get_meter_data(self):
        return {k: v.value()[0] for k, v in self.meters.items()}


def _smooth_l1_loss(x, t, in_weight, sigma):
    sigma2 = sigma ** 2
    diff = in_weight * (x - t)
    abs_diff = diff.abs()
    flag = (abs_diff.data < (1. / sigma2)).float()
    flag = Variable(flag)
    y = (flag * (sigma2 / 2.) * (diff ** 2) +
         (1 - flag) * (abs_diff - 0.5 / sigma2))
    return y.sum()


def _fast_rcnn_loc_loss(pred_loc, gt_loc, gt_label, sigma):
    in_weight = t.zeros(gt_loc.shape).cuda()
    # Localization loss is calculated only for positive rois.
    # NOTE:  unlike origin implementation, 
    # we don't need inside_weight and outside_weight, they can calculate by gt_label
    in_weight[(gt_label > 0).view(-1, 1).expand_as(in_weight).cuda()] = 1
    loc_loss = _smooth_l1_loss(pred_loc, gt_loc, Variable(in_weight), sigma)
    # Normalize by total number of negtive and positive rois.
    loc_loss /= (gt_label >= 0).sum().float()  # ignore gt_label==-1 for rpn_loss
    return loc_loss

def _LocNet_loss(Tx, Ty, Px, Py, gt_label, sigma):
    '''
    Args: 
        Tx, Ty : ground truth value for all points in all boxes. 
                 shape of (R, M) of which R is the number of boxes used in \
                 Head part and M is the number of parts along x and y axis of a box.
        Px, Py : predicted value of Tx and Ty, shape of (R, M)
        gt_label : class id of the box, 0 means background. shape of (R)
    '''
    # print("Tx.shape ", Tx.shape)
    # print("Ty shape ", Ty.shape)
    # print("Tx max ", t.max(Tx))
    # print("Tx min", t.min(Tx))

    s = t.sum(Tx * t.log(Px), dim=1) + t.sum((1-Tx) * t.log(1-Px), dim=1) + t.sum(Ty * t.log(Py), dim=1) + t.sum((1-Ty) * t.log(1-Py), dim=1)   
    s = (-1) * s

    # Localization loss is calculated only for positive rois.
    in_weight = t.zeros(s.shape).cuda()
    for i in range(len(gt_label)):
        if gt_label[i]>0:
            in_weight[i] = 1
    in_weight = Variable(in_weight)

    result = sigma * (in_weight * s).sum()

    return result