maskrcnn_SSA

main(%input_0: (1, 3, 398, 224, fp32)(Tensor),
     %input_1: (1, 3, fp32)(Tensor)) {
  block0() {
    %185: (1, 3, 398, 224, fp32)(Tensor) = sub(x=%input_0, y=[[[103.52999877929688]], [[116.27999877929688]], [[123.67500305175781]]], name="185")
    %input.1: (1, 3, 398, 224, fp32)(Tensor) = real_div(x=%185, y=[[[57.375]], [[57.119998931884766]], [[58.39500045776367]]], name="input.1")
    %input0.1: (1, 16, 199, 112, fp32)(Tensor) = conv(x=%input.1, weight=%151, bias=%150, strides=[2, 2], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=1, name="input0.1")
    %192: (1, 16, 199, 112, fp32)(Tensor) = relu(x=%input0.1, name="192")
    %input1.1: (1, 16, 199, 112, fp32)(Tensor) = conv(x=%192, weight=%147, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=16, name="input1.1")
    %x.1: (1, 16, 199, 112, fp32)(Tensor) = conv(x=%input1.1, weight=%145, bias=%144, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x.1")
    %input2.1: (1, 16, 199, 112, fp32)(Tensor) = add(x=%x.1, y=%192, name="input2.1")
    %input3.1: (1, 64, 199, 112, fp32)(Tensor) = conv(x=%input2.1, weight=%143, bias=%142, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input3.1")
    %209: (1, 64, 199, 112, fp32)(Tensor) = relu(x=%input3.1, name="209")
    %input4.1: (1, 64, 100, 56, fp32)(Tensor) = conv(x=%209, weight=%141, strides=[2, 2], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=64, name="input4.1")
    %input5.1: (1, 32, 100, 56, fp32)(Tensor) = conv(x=%input4.1, weight=%140, bias=%139, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input5.1")
    %input6.1: (1, 64, 100, 56, fp32)(Tensor) = conv(x=%input5.1, weight=%138, bias=%137, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input6.1")
    %225: (1, 64, 100, 56, fp32)(Tensor) = relu(x=%input6.1, name="225")
    %input7.1: (1, 64, 100, 56, fp32)(Tensor) = conv(x=%225, weight=%136, strides=[1, 1], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=64, name="input7.1")
    %x0.1: (1, 32, 100, 56, fp32)(Tensor) = conv(x=%input7.1, weight=%135, bias=%134, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x0.1")
    %input8.1: (1, 32, 100, 56, fp32)(Tensor) = add(x=%x0.1, y=%input5.1, name="input8.1")
    %input9.1: (1, 128, 100, 56, fp32)(Tensor) = conv(x=%input8.1, weight=%133, bias=%132, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input9.1")
    %242: (1, 128, 100, 56, fp32)(Tensor) = relu(x=%input9.1, name="242")
    %input10.1: (1, 128, 50, 28, fp32)(Tensor) = conv(x=%242, weight=%131, strides=[2, 2], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=128, name="input10.1")
    %input11.1: (1, 40, 50, 28, fp32)(Tensor) = conv(x=%input10.1, weight=%130, bias=%129, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input11.1")
    %input12.1: (1, 120, 50, 28, fp32)(Tensor) = conv(x=%input11.1, weight=%128, bias=%127, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input12.1")
    %258: (1, 120, 50, 28, fp32)(Tensor) = relu(x=%input12.1, name="258")
    %input13.1: (1, 120, 50, 28, fp32)(Tensor) = conv(x=%258, weight=%126, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=120, name="input13.1")
    %x1.1: (1, 40, 50, 28, fp32)(Tensor) = conv(x=%input13.1, weight=%125, bias=%124, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x1.1")
    %input14.1: (1, 40, 50, 28, fp32)(Tensor) = add(x=%x1.1, y=%input11.1, name="input14.1")
    %input15.1: (1, 120, 50, 28, fp32)(Tensor) = conv(x=%input14.1, weight=%123, bias=%122, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input15.1")
    %275: (1, 120, 50, 28, fp32)(Tensor) = relu(x=%input15.1, name="275")
    %input16.1: (1, 120, 50, 28, fp32)(Tensor) = conv(x=%275, weight=%121, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=120, name="input16.1")
    %x2.1: (1, 40, 50, 28, fp32)(Tensor) = conv(x=%input16.1, weight=%120, bias=%119, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x2.1")
    %input17.1: (1, 40, 50, 28, fp32)(Tensor) = add(x=%x2.1, y=%input14.1, name="input17.1")
    %input18.1: (1, 120, 50, 28, fp32)(Tensor) = conv(x=%input17.1, weight=%118, bias=%117, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input18.1")
    %292: (1, 120, 50, 28, fp32)(Tensor) = relu(x=%input18.1, name="292")
    %input19.1: (1, 120, 50, 28, fp32)(Tensor) = conv(x=%292, weight=%116, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=120, name="input19.1")
    %x3.1: (1, 40, 50, 28, fp32)(Tensor) = conv(x=%input19.1, weight=%115, bias=%114, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x3.1")
    %input20.1: (1, 40, 50, 28, fp32)(Tensor) = add(x=%x3.1, y=%input17.1, name="input20.1")
    %input21.1: (1, 160, 50, 28, fp32)(Tensor) = conv(x=%input20.1, weight=%113, bias=%112, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input21.1")
    %309: (1, 160, 50, 28, fp32)(Tensor) = relu(x=%input21.1, name="309")
    %input22.1: (1, 160, 25, 14, fp32)(Tensor) = conv(x=%309, weight=%111, strides=[2, 2], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=160, name="input22.1")
    %input23.1: (1, 72, 25, 14, fp32)(Tensor) = conv(x=%input22.1, weight=%110, bias=%109, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input23.1")
    %input24.1: (1, 216, 25, 14, fp32)(Tensor) = conv(x=%input23.1, weight=%108, bias=%107, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input24.1")
    %325: (1, 216, 25, 14, fp32)(Tensor) = relu(x=%input24.1, name="325")
    %input25.1: (1, 216, 25, 14, fp32)(Tensor) = conv(x=%325, weight=%106, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=216, name="input25.1")
    %x4.1: (1, 72, 25, 14, fp32)(Tensor) = conv(x=%input25.1, weight=%105, bias=%104, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x4.1")
    %input26.1: (1, 72, 25, 14, fp32)(Tensor) = add(x=%x4.1, y=%input23.1, name="input26.1")
    %input27.1: (1, 216, 25, 14, fp32)(Tensor) = conv(x=%input26.1, weight=%103, bias=%102, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input27.1")
    %342: (1, 216, 25, 14, fp32)(Tensor) = relu(x=%input27.1, name="342")
    %input28.1: (1, 216, 25, 14, fp32)(Tensor) = conv(x=%342, weight=%101, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=216, name="input28.1")
    %x5.1: (1, 72, 25, 14, fp32)(Tensor) = conv(x=%input28.1, weight=%100, bias=%99, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x5.1")
    %input29.1: (1, 72, 25, 14, fp32)(Tensor) = add(x=%x5.1, y=%input26.1, name="input29.1")
    %input30.1: (1, 216, 25, 14, fp32)(Tensor) = conv(x=%input29.1, weight=%98, bias=%97, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input30.1")
    %359: (1, 216, 25, 14, fp32)(Tensor) = relu(x=%input30.1, name="359")
    %input31.1: (1, 216, 25, 14, fp32)(Tensor) = conv(x=%359, weight=%96, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=216, name="input31.1")
    %x6.1: (1, 72, 25, 14, fp32)(Tensor) = conv(x=%input31.1, weight=%95, bias=%94, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x6.1")
    %input32.1: (1, 72, 25, 14, fp32)(Tensor) = add(x=%x6.1, y=%input29.1, name="input32.1")
    %input33.1: (1, 288, 25, 14, fp32)(Tensor) = conv(x=%input32.1, weight=%93, bias=%92, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input33.1")
    %376: (1, 288, 25, 14, fp32)(Tensor) = relu(x=%input33.1, name="376")
    %input34.1: (1, 288, 25, 14, fp32)(Tensor) = conv(x=%376, weight=%91, strides=[1, 1], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=288, name="input34.1")
    %input35.1: (1, 112, 25, 14, fp32)(Tensor) = conv(x=%input34.1, weight=%90, bias=%89, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input35.1")
    %input36.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%input35.1, weight=%88, bias=%87, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input36.1")
    %392: (1, 448, 25, 14, fp32)(Tensor) = relu(x=%input36.1, name="392")
    %input37.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%392, weight=%86, strides=[1, 1], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=448, name="input37.1")
    %x7.1: (1, 112, 25, 14, fp32)(Tensor) = conv(x=%input37.1, weight=%85, bias=%84, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x7.1")
    %input38.1: (1, 112, 25, 14, fp32)(Tensor) = add(x=%x7.1, y=%input35.1, name="input38.1")
    %input39.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%input38.1, weight=%83, bias=%82, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input39.1")
    %409: (1, 448, 25, 14, fp32)(Tensor) = relu(x=%input39.1, name="409")
    %input40.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%409, weight=%81, strides=[1, 1], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=448, name="input40.1")
    %x8.1: (1, 112, 25, 14, fp32)(Tensor) = conv(x=%input40.1, weight=%80, bias=%79, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x8.1")
    %input41.1: (1, 112, 25, 14, fp32)(Tensor) = add(x=%x8.1, y=%input38.1, name="input41.1")
    %input42.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%input41.1, weight=%78, bias=%77, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input42.1")
    %426: (1, 448, 25, 14, fp32)(Tensor) = relu(x=%input42.1, name="426")
    %input43.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%426, weight=%76, strides=[1, 1], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=448, name="input43.1")
    %x9.1: (1, 112, 25, 14, fp32)(Tensor) = conv(x=%input43.1, weight=%75, bias=%74, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x9.1")
    %input44.1: (1, 112, 25, 14, fp32)(Tensor) = add(x=%x9.1, y=%input41.1, name="input44.1")
    %input45.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%input44.1, weight=%73, bias=%72, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input45.1")
    %443: (1, 448, 25, 14, fp32)(Tensor) = relu(x=%input45.1, name="443")
    %input46.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%443, weight=%71, strides=[1, 1], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=448, name="input46.1")
    %x10.1: (1, 112, 25, 14, fp32)(Tensor) = conv(x=%input46.1, weight=%70, bias=%69, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x10.1")
    %input47.1: (1, 112, 25, 14, fp32)(Tensor) = add(x=%x10.1, y=%input44.1, name="input47.1")
    %input48.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%input47.1, weight=%68, bias=%67, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input48.1")
    %460: (1, 448, 25, 14, fp32)(Tensor) = relu(x=%input48.1, name="460")
    %input49.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%460, weight=%66, strides=[1, 1], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=448, name="input49.1")
    %x11.1: (1, 112, 25, 14, fp32)(Tensor) = conv(x=%input49.1, weight=%65, bias=%64, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x11.1")
    %input50.1: (1, 112, 25, 14, fp32)(Tensor) = add(x=%x11.1, y=%input47.1, name="input50.1")
    %input51.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%input50.1, weight=%63, bias=%62, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input51.1")
    %477: (1, 448, 25, 14, fp32)(Tensor) = relu(x=%input51.1, name="477")
    %input52.1: (1, 448, 25, 14, fp32)(Tensor) = conv(x=%477, weight=%61, strides=[1, 1], pad_type="custom", pad=[2, 2, 2, 2], dilations=[1, 1], groups=448, name="input52.1")
    %x12.1: (1, 112, 25, 14, fp32)(Tensor) = conv(x=%input52.1, weight=%60, bias=%59, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x12.1")
    %input53.1: (1, 112, 25, 14, fp32)(Tensor) = add(x=%x12.1, y=%input50.1, name="input53.1")
    %scores.1: (1, 15, 25, 14, fp32)(Tensor) = conv(x=%input53.1, weight=%58, bias=%57, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="scores.1")
    %bbox_deltas.1: (1, 60, 25, 14, fp32)(Tensor) = conv(x=%input53.1, weight=%56, bias=%55, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="bbox_deltas.1")
    %generateproposals_0:0: (fp32)(Tensor), %generateproposals_0:1: (fp32)(Tensor) = generateproposals(scores=%scores.1, bbox_deltas=%bbox_deltas.1, im_info=%input_1, anchors=%54, spatial_scale=0.0625, rpn_pre_nms_topN=100, rpn_post_nms_topN=10, nms_thresh=0.7, rpn_min_size=0.0, angle_bound_on=True, angle_bound_lo=-180, angle_bound_hi=180, clip_angle_thresh=1.0, legacy_plus_one=False, name="generateproposals_0")
    %roialign_0: (1, 112, 6, 6, fp32)(Tensor) = roialign(features=%input44.1, rois=%generateproposals_0:0, order="NCHW", spatial_scale=0.0625, aligned_height=6, aligned_width=6, sampling_ratio=0, aligned=True, name="roialign_0")
    %input55.1: (1, 448, 6, 6, fp32)(Tensor) = conv(x=%roialign_0, weight=%52, bias=%51, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input55.1")
    %509: (1, 448, 6, 6, fp32)(Tensor) = relu(x=%input55.1, name="509")
    %input56.1: (1, 448, 3, 3, fp32)(Tensor) = conv(x=%509, weight=%50, strides=[2, 2], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=448, name="input56.1")
    %input57.1: (1, 128, 3, 3, fp32)(Tensor) = conv(x=%input56.1, weight=%49, bias=%48, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input57.1")
    %input58.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%input57.1, weight=%47, bias=%46, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input58.1")
    %525: (1, 768, 3, 3, fp32)(Tensor) = relu(x=%input58.1, name="525")
    %input59.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%525, weight=%45, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=768, name="input59.1")
    %x13.1: (1, 128, 3, 3, fp32)(Tensor) = conv(x=%input59.1, weight=%44, bias=%43, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x13.1")
    %input60.1: (1, 128, 3, 3, fp32)(Tensor) = add(x=%x13.1, y=%input57.1, name="input60.1")
    %input61.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%input60.1, weight=%42, bias=%41, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input61.1")
    %542: (1, 768, 3, 3, fp32)(Tensor) = relu(x=%input61.1, name="542")
    %input62.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%542, weight=%40, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=768, name="input62.1")
    %x14.1: (1, 128, 3, 3, fp32)(Tensor) = conv(x=%input62.1, weight=%39, bias=%38, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x14.1")
    %input63.1: (1, 128, 3, 3, fp32)(Tensor) = add(x=%x14.1, y=%input60.1, name="input63.1")
    %input64.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%input63.1, weight=%37, bias=%36, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input64.1")
    %559: (1, 768, 3, 3, fp32)(Tensor) = relu(x=%input64.1, name="559")
    %input65.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%559, weight=%35, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=768, name="input65.1")
    %x15.1: (1, 160, 3, 3, fp32)(Tensor) = conv(x=%input65.1, weight=%34, bias=%33, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x15.1")
    %x16.1: (1, 160, 1, 1, fp32)(Tensor) = reduce_mean(x=%x15.1, axes=[-2, -1], keep_dims=True, name="x16.1")
    %input66.1: (1, 160, fp32)(Tensor) = reshape(x=%x16.1, shape=[1, 160], name="input66.1")
    %573: (1, 2, fp32)(Tensor) = linear(x=%input66.1, weight=%32, bias=[-0.009787099435925484, 0.009667695499956608], name="573")
    %574: (1, 4, fp32)(Tensor) = linear(x=%input66.1, weight=%30, bias=[0.004816785920411348, 0.012010490521788597, -0.026378868147730827, 0.1014462262392044], name="574")
    %class_prob.1: (1, 2, fp32)(Tensor) = softmax(x=%573, axis=-1, name="class_prob.1")
    %tensor.1: (1,fp32)^(Tensor) = concat(values=(%generateproposals_0:0), axis=0, interleave=False, name="tensor.1")
    %bboxtransform_0:0: (1, 4, fp32)(Tensor), %bboxtransform_0:1: (1,fp32)(Tensor) = bboxtransform(roi_in=%tensor.1, delta_in=%574, iminfo_in=%input_1, weights=[10.0, 10.0, 5.0, 5.0], apply_scale=True, rotated=False, angle_bound_on=True, angle_bound_lo=-180, angle_bound_hi=180, clip_angle_thresh=1.0, legacy_plus_one=False, name="bboxtransform_0")
    %boxwithnmslimit_0:0: (fp32)(Tensor), %boxwithnmslimit_0:1: (fp32)(Tensor), %boxwithnmslimit_0:2: (fp32)(Tensor), %boxwithnmslimit_0:3: (fp32)(Tensor), %boxwithnmslimit_0:4: (fp32)(Tensor), %boxwithnmslimit_0:5: (fp32)(Tensor) = boxwithnmslimit(tscores=%class_prob.1, tboxes=%bboxtransform_0:0, tbatch_splits=%bboxtransform_0:1, score_thres=0.05, nms_thres=0.5, detections_per_im=100, soft_nms_enabled=False, soft_nms_method_str="linear", soft_nms_sigma=0.5, soft_nms_min_score_thres=0.001, rotated=False, cls_agnostic_bbox_reg=True, input_boxes_include_bg_cls=False, output_classes_include_bg_cls=False, legacy_plus_one=False, name="boxwithnmslimit_0")
    %roialign_1: (1, 112, 6, 6, fp32)(Tensor) = roialign(features=%input44.1, rois=%boxwithnmslimit_0:1, order="NCHW", spatial_scale=0.0625, aligned_height=6, aligned_width=6, sampling_ratio=0, aligned=True, name="roialign_1")
    %input68.1: (1, 448, 6, 6, fp32)(Tensor) = conv(x=%roialign_1, weight=%27, bias=%26, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input68.1")
    %593: (1, 448, 6, 6, fp32)(Tensor) = relu(x=%input68.1, name="593")
    %input69.1: (1, 448, 3, 3, fp32)(Tensor) = conv(x=%593, weight=%25, strides=[2, 2], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=448, name="input69.1")
    %input70.1: (1, 128, 3, 3, fp32)(Tensor) = conv(x=%input69.1, weight=%24, bias=%23, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input70.1")
    %input71.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%input70.1, weight=%22, bias=%21, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input71.1")
    %609: (1, 768, 3, 3, fp32)(Tensor) = relu(x=%input71.1, name="609")
    %input72.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%609, weight=%20, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=768, name="input72.1")
    %x17.1: (1, 128, 3, 3, fp32)(Tensor) = conv(x=%input72.1, weight=%19, bias=%18, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x17.1")
    %input73.1: (1, 128, 3, 3, fp32)(Tensor) = add(x=%x17.1, y=%input70.1, name="input73.1")
    %input74.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%input73.1, weight=%17, bias=%16, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input74.1")
    %626: (1, 768, 3, 3, fp32)(Tensor) = relu(x=%input74.1, name="626")
    %input75.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%626, weight=%15, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=768, name="input75.1")
    %x18.1: (1, 128, 3, 3, fp32)(Tensor) = conv(x=%input75.1, weight=%14, bias=%13, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="x18.1")
    %input76.1: (1, 128, 3, 3, fp32)(Tensor) = add(x=%x18.1, y=%input73.1, name="input76.1")
    %input77.1: (1, 768, 3, 3, fp32)(Tensor) = conv(x=%input76.1, weight=%12, bias=%11, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input77.1")
    %643: (1, 768, 3, 3, fp32)(Tensor) = relu(x=%input77.1, name="643")
    %input78.1: (1, 768, 6, 6, fp32)(Tensor) = upsample_nearest_neighbor(x=%643, scale_factor_height=2.0, scale_factor_width=2.0, name="input78.1")
    %input79.1: (1, 768, 6, 6, fp32)(Tensor) = conv(x=%input78.1, weight=%10, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=768, name="input79.1")
    %input80.1: (1, 128, 6, 6, fp32)(Tensor) = conv(x=%input79.1, weight=%9, bias=%8, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input80.1")
    %input81.1: (1, 384, 6, 6, fp32)(Tensor) = conv(x=%input80.1, weight=%7, bias=%6, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input81.1")
    %661: (1, 384, 6, 6, fp32)(Tensor) = relu(x=%input81.1, name="661")
    %input82.1: (1, 384, 12, 12, fp32)(Tensor) = upsample_nearest_neighbor(x=%661, scale_factor_height=2.0, scale_factor_width=2.0, name="input82.1")
    %input83.1: (1, 384, 12, 12, fp32)(Tensor) = conv(x=%input82.1, weight=%5, strides=[1, 1], pad_type="custom", pad=[1, 1, 1, 1], dilations=[1, 1], groups=384, name="input83.1")
    %input84.1: (1, 64, 12, 12, fp32)(Tensor) = conv(x=%input83.1, weight=%4, bias=%3, strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="input84.1")
    %pred_mask_logits.1: (1, 1, 12, 12, fp32)(Tensor) = conv(x=%input84.1, weight=%2, bias=[0.09805035591125488], strides=[1, 1], pad_type="custom", pad=[0, 0, 0, 0], dilations=[1, 1], groups=1, name="pred_mask_logits.1")
    %679: (1, 1, 12, 12, fp32)(Tensor) = sigmoid(x=%pred_mask_logits.1, name="679")
  } -> (%boxwithnmslimit_0:1, %boxwithnmslimit_0:0, %boxwithnmslimit_0:2, %679)
}