Faster RCNN roidb.py

1. self.roidb返回一個列表，列表元素爲字典，每張圖片對應一個字典（包括flipped的圖像）

2. prepare_roidb 函數

def prepare_roidb(imdb):
    """Enrich the imdb's roidb by adding some derived quantities that
    are useful for training. This function precomputes the maximum
    overlap, taken over ground-truth boxes, between each ROI and
    each ground-truth box. The class with maximum overlap is also
    recorded.
    該函數主要是用來準備imdb的roidb，主要工作是給roidb中的字典添加一些屬性，比如‘image’。
    簡單地來說，imdb就是圖像數據庫，roidb就是圖像中的Region Of Intrest 數據庫
    """
    sizes = [PIL.Image.open(imdb.image_path_at(i)).size
             for i in xrange(imdb.num_images)]
    # 當在‘Stage 2 Fast R-CNN, init from stage 2 RPN R-CNN model’階段中，roidb由rpn_roidb（）
    # 方法生成，其中的沒一張圖像的box不僅僅只有gtbox，還包括rpn_file裏面的box。
    roidb = imdb.roidb 

    # print '~~~~~~~~~~~~~~imdb.image_index: {}'.format(imdb.image_index)
    # print '~~~~~~~~~~~~~~imdb.image_index: {} {}'.format(imdb.image_index[0], imdb.image_index[5011]) # 000005 000005 說明image_index是重複的兩端合在一起的，即驗證了append_flipped_images()中的self._image_index = self._image_index * 2
    # print '~~~~~~roidb[0]: {}'.format(roidb[0])
    # print '~~~~~~len(imdb.image_index): {}'.format(len(imdb.image_index)) #10022
    for i in xrange(len(imdb.image_index)):
        roidb[i]['image'] = imdb.image_path_at(i)
        roidb[i]['width'] = sizes[i][0]
        roidb[i]['height'] = sizes[i][1]
        # need gt_overlaps as a dense array for argmax
        gt_overlaps = roidb[i]['gt_overlaps'].toarray()
        # max overlap with gt over classes (columns)
        max_overlaps = gt_overlaps.max(axis=1)
        # gt class that had the max overlap 返回最大值的aixs=1軸上的座標，在這裏它同時代表某一類
        max_classes = gt_overlaps.argmax(axis=1)
        roidb[i]['max_classes'] = max_classes
        roidb[i]['max_overlaps'] = max_overlaps
        # sanity checks
        # max overlap of 0 => class should be zero (background)
        #檢查背景類
        zero_inds = np.where(max_overlaps == 0)[0]
        assert all(max_classes[zero_inds] == 0)
        # max overlap > 0 => class should not be zero (must be a fg class)
        nonzero_inds = np.where(max_overlaps > 0)[0]
        assert all(max_classes[nonzero_inds] != 0)
        # print '~~~~~~~~~~~~~~~~zero_inds: {}'.format(zero_inds)
    # print '~~~~~~~~roidb[0]["gt_overlaps"]: {}'.format(roidb[0]['gt_overlaps'].toarray())
    # print '~~~~~~~~roidb[1]["gt_overlaps"]: {}'.format(roidb[1]['gt_overlaps'].toarray())
    # print '~~~~~~~~roidb[2]["gt_overlaps"]: {}'.format(roidb[2]['gt_overlaps'].toarray())
    # print '~~~~~~~~roidb[3]["gt_overlaps"]: {}'.format(roidb[3]['gt_overlaps'].toarray())

print ‘~~~~roidb[3][“gt_overlaps”]: {}’.format(roidb[3][‘gt_overlaps’].toarray())等測試代碼在stage 1 時的輸出如下：

{'boxes': array([[262, 210, 323, 338],
                [164, 263, 252, 371],
                [240, 193, 294, 298]], dtype=uint16), 'gt_classes': array([9, 9, 9], dtype=int32), 'gt_overlaps': <3x21 sparse matrix of type '<type 'numpy.float32'>'
    with 3 stored elements in Compressed Sparse Row format>, 'seg_areas': array([ 7998.,  9701.,  5830.], dtype=float32), 'flipped': False}
## 'gt_overlaps': <3x21 sparse matrix of type '<type 'numpy.float32'>' with 3 stored elements in Compressed Sparse Row format> 21代表21類，
#with 3 stored elements in Compressed Sparse Row format> 說明該sparse matrix裏面只存儲了3個非0值


#roidb[0]
{'gt_classes': array([9, 9, 9], dtype=int32), 'max_classes': array([9, 9, 9]), 'image': '/home/sam/WORKSPACE/py-faster-rcnn/data/VOCdevkit2007/VOC2007/JPEGImages/000005.jpg', 'flipped': False, 'width': 500, 
'boxes': array([[262, 210, 323, 338],
                [164, 263, 252, 371],
                [240, 193, 294, 298]], dtype=uint16), 'max_overlaps': array([ 1.,  1.,  1.], dtype=float32), 'height': 375, 'seg_areas': array([ 7998.,  9701.,  5830.], dtype=float32), 'gt_overlaps': <3x21 sparse matrix of type '<type 'numpy.float32'>'
    with 3 stored elements in Compressed Sparse Row format>}


#roidb[1]
{'gt_classes': array([7], dtype=int32), 'max_classes': array([7]), 'image': '/home/sam/WORKSPACE/py-faster-rcnn/data/VOCdevkit2007/VOC2007/JPEGImages/000007.jpg', 'flipped': False, 'width': 500, 
    'boxes': array([[140,  49, 499, 329]], dtype=uint16), 'max_overlaps': array([ 1.], dtype=float32), 'height': 333, 'seg_areas': array([ 101160.], dtype=float32), 'gt_overlaps': <1x21 sparse matrix of type '<type 'numpy.float32'>'
    with 1 stored elements in Compressed Sparse Row format>}


#roidb[2]
{'gt_classes': array([13, 15, 15, 15], dtype=int32), 'max_classes': array([13, 15, 15, 15]), 'image': '/home/sam/WORKSPACE/py-faster-rcnn/data/VOCdevkit2007/VOC2007/JPEGImages/000009.jpg', 'flipped': False, 'width': 500, 
'boxes': array([[ 68, 171, 269, 329],
                [149, 140, 228, 283],
                [284, 200, 326, 330],
                [257, 197, 296, 328]], dtype=uint16), 'max_overlaps': array([ 1.,  1.,  1.,  1.], dtype=float32), 'height': 375, 'seg_areas': array([ 32118.,  11520.,   5633.,   5280.], dtype=float32), 'gt_overlaps': <4x21 sparse matrix of type '<type 'numpy.float32'>'
    with 4 stored elements in Compressed Sparse Row format>}


#roidb[3]
{'gt_classes': array([7], dtype=int32), 'max_classes': array([7]), 'image': '/home/sam/WORKSPACE/py-faster-rcnn/data/VOCdevkit2007/VOC2007/JPEGImages/000012.jpg', 'flipped': False, 'width': 500, 
'boxes': array([[155,  96, 350, 269]], dtype=uint16), 'max_overlaps': array([ 1.], dtype=float32), 'height': 333, 'seg_areas': array([ 34104.], dtype=float32), 'gt_overlaps': <1x21 sparse matrix of type '<type 'numpy.float32'>'
    with 1 stored elements in Compressed Sparse Row format>}

#roidb[0]["gt_overlaps"]: 
[[ 0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0. 0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0. 0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0. 0.  0.  0.]]

#roidb[1]["gt_overlaps"]: 
[[ 0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0. 0.  0.  0.]]

#roidb[2]["gt_overlaps"]: 
[[ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.]
 [ 0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.]]

#roidb[3]["gt_overlaps"]: 
[[ 0.  0.  0.  0.  0.  0.  0.  1.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.  0.]]

def add_bbox_regression_targets(roidb): 添加bbox的迴歸目標（給roidb裏面的字典添加了’bbox_targets’屬性），同時根據cfg文件來處理bbox_targets的mean stds 問題

def add_bbox_regression_targets(roidb):
    """Add information needed to train bounding-box regressors."""
    assert len(roidb) > 0
    assert 'max_classes' in roidb[0], 'Did you call prepare_roidb first?'

    num_images = len(roidb)
    # Infer number of classes from the number of columns in gt_overlaps
    num_classes = roidb[0]['gt_overlaps'].shape[1]
    for im_i in xrange(num_images):
        rois = roidb[im_i]['boxes']

        # 當在‘Stage 2 Fast R-CNN, init from stage 2 RPN R-CNN model’階段中，roidb由rpn_roidb（）
        # 方法生成，其中的沒一張圖像的box不僅僅只有gtbox，還包括rpn_file裏面的box。
        max_overlaps = roidb[im_i]['max_overlaps']
        max_classes = roidb[im_i]['max_classes']
        roidb[im_i]['bbox_targets'] = \
                _compute_targets(rois, max_overlaps, max_classes)

    if cfg.TRAIN.BBOX_NORMALIZE_TARGETS_PRECOMPUTED:
        # Use fixed / precomputed "means" and "stds" instead of empirical values
        means = np.tile(
                np.array(cfg.TRAIN.BBOX_NORMALIZE_MEANS), (num_classes, 1))
        stds = np.tile(
                np.array(cfg.TRAIN.BBOX_NORMALIZE_STDS), (num_classes, 1))
    else:
        # Compute values needed for means and stds
        # var(x) = E(x^2) - E(x)^2
        # 針對每個類計算mean 和 std， 因爲需要訓練class-specific regressors
        class_counts = np.zeros((num_classes, 1)) + cfg.EPS
        sums = np.zeros((num_classes, 4))
        squared_sums = np.zeros((num_classes, 4))
        for im_i in xrange(num_images):
            targets = roidb[im_i]['bbox_targets']
            for cls in xrange(1, num_classes):
                cls_inds = np.where(targets[:, 0] == cls)[0]
                if cls_inds.size > 0:
                    class_counts[cls] += cls_inds.size
                    sums[cls, :] += targets[cls_inds, 1:].sum(axis=0)
                    squared_sums[cls, :] += \
                            (targets[cls_inds, 1:] ** 2).sum(axis=0)

        means = sums / class_counts
        stds = np.sqrt(squared_sums / class_counts - means ** 2)

    print 'bbox target means:'
    print means # 輸出21類的bbox_targets的4個offset的均值
    print means[1:, :].mean(axis=0) # ignore bg class
    print 'bbox target stdevs:'
    print stds # 輸出21類的bbox_targets的4個offset的std，每一類爲1行
    print stds[1:, :].mean(axis=0) # ignore bg class

    # Normalize targets
    if cfg.TRAIN.BBOX_NORMALIZE_TARGETS:
        print "Normalizing targets"
        for im_i in xrange(num_images):
            targets = roidb[im_i]['bbox_targets']
            for cls in xrange(1, num_classes):
                cls_inds = np.where(targets[:, 0] == cls)[0]
                roidb[im_i]['bbox_targets'][cls_inds, 1:] -= means[cls, :]
                roidb[im_i]['bbox_targets'][cls_inds, 1:] /= stds[cls, :]
    else:
        print "NOT normalizing targets"

    # These values will be needed for making predictions
    # (the predicts will need to be unnormalized and uncentered)
    return means.ravel(), stds.ravel()

def _compute_targets(rois, overlaps, labels)
roidb中鍵‘boxes’所對應的值爲左上、右下兩點的座標，其形式爲（w1, h1, w2, h2）

def _compute_targets(rois, overlaps, labels):
    """Compute bounding-box regression targets for an image."""
    # Indices of ground-truth ROIs
    gt_inds = np.where(overlaps == 1)[0]
    if len(gt_inds) == 0:
        # Bail if the image has no ground-truth ROIs
        return np.zeros((rois.shape[0], 5), dtype=np.float32)
    # Indices of examples for which we try to make predictions
    ex_inds = np.where(overlaps >= cfg.TRAIN.BBOX_THRESH)[0]

    # Get IoU overlap between each ex ROI and gt ROI
    # 調用bbox.pyx文件中的bbox_overlaps函數計算RPN Proposals 與 Ground-Truth Box的IOU
    # overlap;IOU overlap爲Proposal與gt box的面積交集 除以 面積的並集。
    ex_gt_overlaps = bbox_overlaps(
        np.ascontiguousarray(rois[ex_inds, :], dtype=np.float),
        np.ascontiguousarray(rois[gt_inds, :], dtype=np.float))

    # Find which gt ROI each ex ROI has max overlap with:
    # this will be the ex ROI's gt target
    gt_assignment = ex_gt_overlaps.argmax(axis=1)

    # 這裏gt_assignment的size比gt_inds的size大，所以返回gt_rois的size和ex_rois的size一樣大，
    #即爲ex_rois中的每個roi都assign了一個gt，可見python的靈活強大
    gt_rois = rois[gt_inds[gt_assignment], :]
    ex_rois = rois[ex_inds, :]

    # 調用bbox_transform.py文件中的bbox_transform函數。target返回圖片中每個box的結果，
    #被選爲（P，G）pair的proposals包含其中，那些不滿足overlaps >= cfg.TRAIN.BBOX_THRESH 
    #的proposal也包含其中，只不過相應的值爲0.bbox_transform的計算參考論文RCNN
    targets = np.zeros((rois.shape[0], 5), dtype=np.float32)
    targets[ex_inds, 0] = labels[ex_inds]
    targets[ex_inds, 1:] = bbox_transform(ex_rois, gt_rois)
    return targets

bbox.pyx文件中的 def bbox_overlaps(np.ndarray[DTYPE_t, ndim=2] boxes,np.ndarray[DTYPE_t, ndim=2] query_boxes):
返回： overlaps: (N, K) ndarray of overlap between boxes and query_boxes

def bbox_overlaps(
        np.ndarray[DTYPE_t, ndim=2] boxes,
        np.ndarray[DTYPE_t, ndim=2] query_boxes):
    """
    Parameters
    ----------
    boxes: (N, 4) ndarray of float
    query_boxes: (K, 4) ndarray of float
    Returns
    -------
    overlaps: (N, K) ndarray of overlap between boxes and query_boxes
    """
    cdef unsigned int N = boxes.shape[0]
    cdef unsigned int K = query_boxes.shape[0]
    cdef np.ndarray[DTYPE_t, ndim=2] overlaps = np.zeros((N, K), dtype=DTYPE)
    cdef DTYPE_t iw, ih, box_area
    cdef DTYPE_t ua
    cdef unsigned int k, n
    for k in range(K):
        box_area = (
            (query_boxes[k, 2] - query_boxes[k, 0] + 1) *
            (query_boxes[k, 3] - query_boxes[k, 1] + 1)
        )
        for n in range(N):
            iw = (
                min(boxes[n, 2], query_boxes[k, 2]) -
                max(boxes[n, 0], query_boxes[k, 0]) + 1
            )
            if iw > 0:
                ih = (
                    min(boxes[n, 3], query_boxes[k, 3]) -
                    max(boxes[n, 1], query_boxes[k, 1]) + 1
                )
                if ih > 0:
                    ua = float(
                        (boxes[n, 2] - boxes[n, 0] + 1) *
                        (boxes[n, 3] - boxes[n, 1] + 1) +
                        box_area - iw * ih
                    )
                    overlaps[n, k] = iw * ih / ua
    return overlaps

Faster RCNN roidb.py

CORS error 但是 status code 是200 OK

壓縮上傳的GPU數據的方案

使用skopeo同步鏡像

Faster RCNN roi_pooling_layer.cpp roi_pooling_layer.cu

Faster RCNN pascal_voc.py

Faster RCNN layer.py

Faster RCNN train.py

Faster RCNN imdb.py

Mac下配置sublime實現LaTeX

https://yachay.unat.edu.pe/blog/index.php?comment_area=format_blog&comment_component=blog&comment_co

linux以太網驅動總結