Skip to content

chapter13-15 代码接口更新 #64

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
Liiiii1234 wants to merge 5 commits into
base: master
Choose a base branch
from
Open

chapter13-15 代码接口更新 #64

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
60bbf78
Migrate ops to mint interface
Liiiii1234 Dec 21, 2025
347c2eb
Migrate ops to mint interface
Liiiii1234 Dec 21, 2025
0e10b62
Migrate ops to mint interface
Liiiii1234 Dec 21, 2025
a98fab7
Migrate ops to mint interface in chapter_13_chapter_computer-vision
Liiiii1234 Dec 22, 2025
2682219
remove build metadata and kernel-meta files
Mar 19, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
101 changes: 52 additions & 49 deletions chapter_13_chapter_computer-vision/boundingbox.ipynb
View file
Open in desktop

Large diffs are not rendered by default.

801 changes: 659 additions & 142 deletions chapter_13_chapter_computer-vision/fcn.ipynb
View file
Open in desktop

Large diffs are not rendered by default.

49 changes: 49 additions & 0 deletions chapter_13_chapter_computer-vision/ge_check_op.json
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,49 @@
{
"graph_id": 14,
"op": [
{
"error_type": "infer_shape_error",
"input0": {
"data_type": "DT_INT32",
"layout": "NCHW",
"shape": [
4
]
},
"input1": {
"data_type": "DT_FLOAT",
"layout": "NCHW",
"shape": [
21,
21,
64,
64
]
},
"input2": {
"data_type": "DT_FLOAT",
"layout": "NCHW",
"shape": [
32,
21,
10,
15
]
},
"name": "Conv2DBackpropInput15",
"output0": {
"data_type": "DT_FLOAT",
"layout": "NCHW",
"shape": [
32,
21,
288,
448
]
},
"reason": "InferShapeFailed!",
"type": "Conv2DBackpropInput"
}
],
"session_id": 14
}
284 changes: 162 additions & 122 deletions chapter_13_chapter_computer-vision/multiscale-object-detection.ipynb
View file
Open in desktop

Large diffs are not rendered by default.

121 changes: 80 additions & 41 deletions chapter_13_chapter_computer-vision/ssd.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,22 @@
"execution_count": 1,
"id": "e6901d6d",
"metadata": {},
"outputs": [],
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"[WARNING] CORE(503527,ffff9d2cb640,python):2025-12-22-00:20:08.955.715 [mindspore/core/utils/ms_context.cc:533] GetJitLevel] Set jit level to O2 for rank table startup method.\n"
]
}
],
"source": [
"%matplotlib inline\n",
"import mindspore\n",
"from mindspore import nn\n",
"from mindspore import mint\n",
"import sys\n",
"sys.path.insert(0, \"..\")\n",
"from d2l import mindspore as d2l\n",
"\n",
"\n",
Expand Down Expand Up @@ -90,8 +101,8 @@
"def forward(x, block):\n",
" return block(x)\n",
"\n",
"Y1 = forward(d2l.zeros((2, 8, 20, 20)), cls_predictor(8, 5, 10))\n",
"Y2 = forward(d2l.zeros((2, 16, 10, 10)), cls_predictor(16, 3, 10))\n",
"Y1 = forward(mint.zeros((2, 8, 20, 20)), cls_predictor(8, 5, 10))\n",
"Y2 = forward(mint.zeros((2, 16, 10, 10)), cls_predictor(16, 3, 10))\n",
"Y1.shape, Y2.shape"
]
},
Expand All @@ -103,10 +114,10 @@
"outputs": [],
"source": [
"def flatten_pred(pred):\n",
" return d2l.flatten(pred.permute(0, 2, 3, 1)) # flatten不改变0轴的size\n",
" return mint.flatten(pred.permute(0, 2, 3, 1), start_dim=1) # flatten不改变0轴的size\n",
"\n",
"def concat_preds(preds):\n",
" return d2l.concat([flatten_pred(p) for p in preds], axis=1)"
" return mint.concat([flatten_pred(p) for p in preds], dim=1)"
]
},
{
Expand Down Expand Up @@ -175,7 +186,7 @@
}
],
"source": [
"forward(d2l.zeros((2, 3, 20, 20)), down_sample_blk(3, 10)).shape"
"forward(mint.zeros((2, 3, 20, 20)), down_sample_blk(3, 10)).shape"
]
},
{
Expand Down Expand Up @@ -211,7 +222,7 @@
" blk.append(down_sample_blk(num_filters[i], num_filters[i+1]))\n",
" return nn.SequentialCell(*blk)\n",
"\n",
"forward(d2l.zeros((2, 3, 256, 256)), base_net()).shape"
"forward(mint.zeros((2, 3, 256, 256)), base_net()).shape"
]
},
{
Expand Down Expand Up @@ -296,7 +307,7 @@
" X, anchors[i], cls_preds[i], bbox_preds[i] = blk_forward(\n",
" X, getattr(self, f'blk_{i}'), sizes[i], ratios[i],\n",
" getattr(self, f'cls_{i}'), getattr(self, f'bbox_{i}'))\n",
" anchors = d2l.concat(anchors, axis=1)\n",
" anchors = mint.concat(anchors, dim=1)\n",
" cls_preds = concat_preds(cls_preds)\n",
" cls_preds = cls_preds.reshape(\n",
" cls_preds.shape[0], -1, self.num_classes + 1)\n",
Expand All @@ -322,7 +333,7 @@
],
"source": [
"net = TinySSD(num_classes=1)\n",
"X = d2l.zeros((32, 3, 256, 256))\n",
"X = mint.zeros((32, 3, 256, 256))\n",
"anchors, cls_preds, bbox_preds = net(X)\n",
"\n",
"print('output anchors:', anchors.shape)\n",
Expand Down Expand Up @@ -357,13 +368,18 @@
"output_type": "stream",
"text": [
"read 1000 training examples\n",
"read 100 validation examples\n"
"read 100 validation examples\n",
"250\n"
]
}
],
"source": [
"batch_size = 32\n",
"train_iter, _ = d2l.load_data_bananas(batch_size) # "
"# 将 Batch Size 设为 4\n",
"# 后续的 multibox_target 函数包含大量基于 CPU 的 Python 循环操作。如果 Batch Size 过大，CPU 计算压力过大导致流水线阻塞，NPU 长期空闲，程序会表现为卡死。\n",
"# 改小 Batch Size 可显著减少单次迭代的 CPU 负担，快速跑通代码。\n",
"batch_size = 4\n",
"train_iter, _ = d2l.load_data_bananas(batch_size) # \n",
"print(train_iter.get_dataset_size())"
]
},
{
Expand Down Expand Up @@ -398,7 +414,7 @@
"def calc_loss(cls_preds, cls_labels, bbox_preds, bbox_labels, bbox_masks):\n",
" batch_size, num_classes = cls_preds.shape[0], cls_preds.shape[2]\n",
" cls = cls_loss(cls_preds.reshape(-1, num_classes),\n",
" cls_labels.reshape(-1)).reshape(batch_size, -1).mean(axis=1)\n",
" cls_labels.reshape(-1).astype(mindspore.int32)).reshape(batch_size, -1).mean(axis=1)\n",
" bbox = bbox_loss(bbox_preds * bbox_masks,\n",
" bbox_labels * bbox_masks).mean(axis=1)\n",
" return cls + bbox"
Expand All @@ -417,7 +433,7 @@
" cls_labels.dtype) == cls_labels).sum())\n",
"\n",
"def bbox_eval(bbox_preds, bbox_labels, bbox_masks):\n",
"# return float((d2l.abs((bbox_labels - bbox_preds) * bbox_masks)).sum())\n",
"# return float((mint.abs((bbox_labels - bbox_preds) * bbox_masks)).sum())\n",
" return float(((bbox_labels - bbox_preds) * bbox_masks).abs().sum())"
]
},
Expand Down Expand Up @@ -493,7 +509,7 @@
" label[:, 1:], anchors)\n",
" bbox_mask = ops.tile((anchors_bbox_map >= 0).float().unsqueeze(-1), (1, 4))\n",
" # 将类标签和分配的边界框坐标初始化为零\n",
" class_labels = ops.zeros(num_anchors, dtype=mindspore.int32)\n",
" class_labels = ops.zeros(num_anchors, dtype=mindspore.int64)\n",
" assigned_bb = ops.zeros((num_anchors, 4), dtype=mindspore.float32)\n",
" # 使用真实边界框来标记锚框的类别。\n",
" # 如果一个锚框没有被分配，标记其为背景（值为零）\n",
Expand Down Expand Up @@ -525,7 +541,7 @@
"name": "stdout",
"output_type": "stream",
"text": [
"(32, 3, 256, 256) (32, 1, 5)\n"
"(4, 3, 256, 256) (4, 1, 5)\n"
]
}
],
Expand All @@ -547,8 +563,8 @@
"output_type": "stream",
"text": [
"output anchors: (1, 5444, 4)\n",
"output class preds: (32, 5444, 2)\n",
"output bbox preds: (32, 21776)\n"
"output class preds: (4, 5444, 2)\n",
"output bbox preds: (4, 21776)\n"
]
}
],
Expand All @@ -567,21 +583,13 @@
"scrolled": true
},
"outputs": [
{
"name": "stderr",
"output_type": "stream",
"text": [
"[WARNING] KERNEL(269803,2aecc71afd80,python):2023-02-23-20:56:28.763.896 [mindspore/ccsrc/plugin/device/gpu/kernel/cuda_impl/cuda_class/cuda_class_common.h:51] CalShapesSizeInBytes] For 'Argmax', the shapes[0] is ( )\n",
"[WARNING] KERNEL(269803,2aecc71afd80,python):2023-02-23-20:56:28.763.965 [mindspore/ccsrc/plugin/device/gpu/kernel/cuda_impl/cuda_class/cuda_class_common.h:51] CalShapesSizeInBytes] For 'Argmax', the shapes[0] is ( )\n"
]
},
{
"name": "stdout",
"output_type": "stream",
"text": [
"bbox_labels: (32, 21776) (32, 21776)\n",
"bbox_masks: (32, 21776) Float32\n",
"cls_labels: (32, 5444) Int32\n"
".bbox_labels: (4, 21776) (4, 21776)\n",
"bbox_masks: (4, 21776) Float32\n",
"cls_labels: (4, 5444) Int64\n"
]
}
],
Expand All @@ -601,7 +609,7 @@
{
"data": {
"text/plain": [
"Tensor(shape=[], dtype=Float32, value= 0.69856)"
"Tensor(shape=[], dtype=Float32, value= 0.705361)"
]
},
"execution_count": 22,
Expand All @@ -616,18 +624,34 @@
},
{
"cell_type": "markdown",
"id": "9c78354a",
"id": "11899268-b5c4-4b96-bfbb-3a372813b9e4",
"metadata": {},
"source": [
"# BUG 直接运行的话，这里会卡住"
"#### 13.7.2.3. 单批次试运行\n",
"从数据集中取出一批数据，完整走一遍全部过程，旨在验证各环节的张量形状和计算逻辑是否正确，只跑一轮就退出，不进行参数更新。"
]
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 26,
"id": "5e0d0317",
"metadata": {},
"outputs": [],
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(4, 3, 256, 256) (4, 1, 5)\n",
"output anchors: (1, 5444, 4)\n",
"output class preds: (4, 5444, 2)\n",
"output bbox preds: (4, 21776)\n",
"bbox_labels: (4, 21776) (4, 21776)\n",
"bbox_masks: (4, 21776) Float32\n",
"cls_labels: (4, 5444) Int64\n",
"[0.713037 0.70012754 0.6992287 0.707439 ]\n"
]
}
],
"source": [
"#forward\n",
"for X, Y in train_iter:\n",
Expand All @@ -650,12 +674,27 @@
},
{
"cell_type": "code",
"execution_count": null,
"execution_count": 25,
"id": "6ef20694",
"metadata": {},
"outputs": [],
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"(4, 3, 256, 256) (4, 1, 5)\n",
"output anchors: (1, 5444, 4)\n",
"output class preds: (4, 5444, 2)\n",
"output bbox preds: (4, 21776)\n",
"bbox_labels: (4, 21776) (4, 21776)\n",
"bbox_masks: (4, 21776) Float32\n",
"cls_labels: (4, 5444) Int64\n",
"[0.7040471 0.70199287 0.7137333 0.69965523]\n"
]
}
],
"source": [
"X, Y = next(train_iter)\n",
"X, Y = next(train_iter.create_tuple_iterator())\n",
"print(X.shape, Y.shape)\n",
"anchors, cls_preds, bbox_preds = net(X)\n",
"print('output anchors:', anchors.shape)\n",
Expand All @@ -677,7 +716,7 @@
"id": "6912bb2d",
"metadata": {},
"source": [
"#### 13.7.2.3. 训练模型"
"#### 13.7.2.4. 训练模型"
]
},
{
Expand Down Expand Up @@ -730,9 +769,9 @@
],
"metadata": {
"kernelspec": {
"display_name": "Python 3 (ipykernel)",
"display_name": "Python 3.10",
"language": "python",
"name": "python3"
"name": "py310"
},
"language_info": {
"codemirror_mode": {
Expand All @@ -744,7 +783,7 @@
"name": "python",
"nbconvert_exporter": "python",
"pygments_lexer": "ipython3",
"version": "3.9.5"
"version": "3.10.14"
},
"toc": {
"base_numbering": 1,
Expand Down
Loading