Add a tool named get_static_model_from_algorithm and deploy guide for l1-norm

configs/pruning/mmcls/dmcp/metafile.yml

@@ -0,0 +1,19 @@
+Models:
+  - Name: dmcp_resnet50_subnet_32xb64
+    In Collection: DMCP
+    Config: configs/pruning/mmcls/dmcp/dmcp_resnet50_subnet_32xb64.py
+    Weights: https://download.openmmlab.com/mmrazor/v1/pruning/dmcp/resnet50/2G/DMCP_R50_2G.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 76.11
+  - Name: dmcp_mbv2_subnet_32xb64
+    In Collection: DMCP
+    Config: configs/pruning/mmcls/dmcp/dmcp_mbv2_subnet_32xb64.py
+    Weights: https://download.openmmlab.com/mmrazor/v1/pruning/dmcp/mobilenetv2/100M/DMCP_MBV2_100M.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 67.22

configs/pruning/mmcls/group_fisher/mobilenet/metafile.yml

@@ -0,0 +1,19 @@
+Models:
+  - Name: group_fisher_act_finetune_mobilenet-v2_8xb32_in1k
+    In Collection: GroupFisher
+    Config: configs/pruning/mmcls/group_fisher/mobilenet/group_fisher_act_finetune_mobilenet-v2_8xb32_in1k.py
+    Weights: https://download.openmmlab.com/mmrazor/v1/pruning/group_fisher/mobilenet/act/group_fisher_act_finetune_mobilenet-v2_8xb32_in1k.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 70.82
+  - Name: group_fisher_flops_finetune_mobilenet-v2_8xb32_in1k
+    In Collection: GroupFisher
+    Config: configs/pruning/mmcls/group_fisher/mobilenet/group_fisher_flops_finetune_mobilenet-v2_8xb32_in1k.py
+    Weights: https://download.openmmlab.com/mmrazor/v1/pruning/group_fisher/mobilenet/flop/group_fisher_flops_finetune_mobilenet-v2_8xb32_in1k.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 70.87

configs/pruning/mmcls/group_fisher/resnet50/metafile.yml

@@ -0,0 +1,19 @@
+Models:
+  - Name: group_fisher_act_finetune_resnet50_8xb32_in1k
+    In Collection: GroupFisher
+    Config: configs/pruning/mmcls/group_fisher/resnet50/group_fisher_act_finetune_resnet50_8xb32_in1k.py
+    Weights: https://download.openmmlab.com/mmrazor/v1/pruning/group_fisher/resnet50/act/group_fisher_act_finetune_resnet50_8xb32_in1k.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 75.22
+  - Name: group_fisher_flops_finetune_resnet50_8xb32_in1k
+    In Collection: GroupFisher
+    Config: configs/pruning/mmcls/group_fisher/resnet50/group_fisher_flops_finetune_resnet50_8xb32_in1k.py
+    Weights: https://download.openmmlab.com/mmrazor/v1/pruning/group_fisher/resnet50/flops/group_fisher_flops_finetune_resnet50_8xb32_in1k.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 75.61

configs/pruning/mmcls/l1-norm/README.md

@@ -18,3 +18,44 @@ We use ItePruneAlgorithm and L1MutableChannelUnit to implement l1-norm pruning.
 | ResNet34_Pruned_C | 73.89 | +0.27 | 3.40    | 7.6%   | 2.02       | 7.3%   | [config](./l1-norm_resnet34_8xb32_in1k_a.py)                                                           | [model](https://openmmlab-share.oss-cn-hangzhou.aliyuncs.com/mmrazor/v1/pruning/l1-norm/l1-norm_resnet34_8xb32_in1k_c.pth) \| [log](https://openmmlab-share.oss-cn-hangzhou.aliyuncs.com/mmrazor/v1/pruning/l1-norm/l1-norm_resnet34_8xb32_in1k_c.json) |
 
 **Note:** There is a different implementation from the original paper. We pruned the layers related to the shortcut with a shared pruning decision, while the original paper pruned them separately in *Pruned C*. This may be why our *Pruned C* outperforms *Prune A* and *Prune B*, while *Pruned C* is worst in the original paper.
+
+## Getting Started
+
+### Prune
+
+```bash
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 PORT=29500 ./tools/dist_train.sh \
+  {prune_config_path}.py 8 --work-dir $WORK_DIR
+```
+
+after the pruning process, you can get a checkpoint file in the work_dir. This checkpoint file including all parameters of the original model. In the next step, we will use the checkpoint to export a pruned checkpoint.
+
+### Get the pruned model
+
+```bash
+python ./tools/pruning/get_static_model_from_algorithm.py \
+  {prune_config_path}.py \
+  {checkpoint_file}.pth \
+  --o {output_folder}
+```
+
+This step will export a pruned checkpoint and a json file which records the pruning structure. This two file will be used to deploy the pruned model.
+
+### Deploy
+
+For a pruned model, you only need to use the pruning deploy config instead of the pretrain config to deploy the pruned version of your model. If you are not fimilar with MMDeploy, please refer to [mmdeploy](https://github.com/open-mmlab/mmdeploy/tree/1.x).
+
+```bash
+python {mmdeploy}/tools/deploy.py \
+    {mmdeploy}/{mmdeploy_config}.py \
+    {pruning_deploy_config}.py \
+    {pruned_checkpoint}.pth \
+    {mmdeploy}/tests/data/tiger.jpeg
+```
+
+### Get the Flops and Parameters of a Pruned Model
+
+```bash
+python ./tools/pruning/get_flops.py \
+    {pruning_deploy_config}.py
+```

configs/pruning/mmcls/l1-norm/l1-norm_resnet34_8xb32_in1k_a_deploy.py

@@ -0,0 +1,57 @@
+#############################################################################
+"""You have to fill these args.
+
+_base_(str): The path to your pretrain config file.
+fix_subnet (Union[dict,str]): The dict store the pruning structure or the
+    json file including it.
+divisor (int): The divisor the make the channel number divisible.
+"""
+
+_base_ = ['mmcls::resnet/resnet34_8xb32_in1k.py']
+un_prune = 1.0
+stage_ratio_1 = 0.7
+stage_ratio_2 = 0.7
+stage_ratio_3 = 0.7
+stage_ratio_4 = un_prune
+
+# the config template of target_pruning_ratio can be got by
+# python ./tools/get_channel_units.py {config_file} --choice
+fix_subnet = {
+    # stage 1
+    'backbone.conv1_(0, 64)_64': un_prune,  # short cut layers
+    'backbone.layer1.0.conv1_(0, 64)_64': stage_ratio_1,
+    'backbone.layer1.1.conv1_(0, 64)_64': stage_ratio_1,
+    'backbone.layer1.2.conv1_(0, 64)_64': un_prune,
+    # stage 2
+    'backbone.layer2.0.conv1_(0, 128)_128': un_prune,
+    'backbone.layer2.0.conv2_(0, 128)_128': un_prune,  # short cut layers
+    'backbone.layer2.1.conv1_(0, 128)_128': stage_ratio_2,
+    'backbone.layer2.2.conv1_(0, 128)_128': stage_ratio_2,
+    'backbone.layer2.3.conv1_(0, 128)_128': un_prune,
+    # stage 3
+    'backbone.layer3.0.conv1_(0, 256)_256': un_prune,
+    'backbone.layer3.0.conv2_(0, 256)_256': un_prune,  # short cut layers
+    'backbone.layer3.1.conv1_(0, 256)_256': stage_ratio_3,
+    'backbone.layer3.2.conv1_(0, 256)_256': stage_ratio_3,
+    'backbone.layer3.3.conv1_(0, 256)_256': stage_ratio_3,
+    'backbone.layer3.4.conv1_(0, 256)_256': stage_ratio_3,
+    'backbone.layer3.5.conv1_(0, 256)_256': un_prune,
+    # stage 4
+    'backbone.layer4.0.conv1_(0, 512)_512': stage_ratio_4,
+    'backbone.layer4.0.conv2_(0, 512)_512': un_prune,  # short cut layers
+    'backbone.layer4.1.conv1_(0, 512)_512': stage_ratio_4,
+    'backbone.layer4.2.conv1_(0, 512)_512': stage_ratio_4
+}
+divisor = 8
+##############################################################################
+
+architecture = _base_.model
+
+model = dict(
+    _delete_=True,
+    _scope_='mmrazor',
+    type='GroupFisherDeploySubModel',
+    architecture=architecture,
+    fix_subnet=fix_subnet,
+    divisor=divisor,
+)

configs/pruning/mmcls/l1-norm/l1-norm_resnet34_8xb32_in1k_b_deploy.py

@@ -0,0 +1,57 @@
+#############################################################################
+"""You have to fill these args.
+
+_base_(str): The path to your pretrain config file.
+fix_subnet (Union[dict,str]): The dict store the pruning structure or the
+    json file including it.
+divisor (int): The divisor the make the channel number divisible.
+"""
+
+_base_ = ['mmcls::resnet/resnet34_8xb32_in1k.py']
+
+un_prune = 1.0
+stage_ratio_1 = 0.5
+stage_ratio_2 = 0.4
+stage_ratio_3 = 0.6
+stage_ratio_4 = un_prune
+
+fix_subnet = {
+    # stage 1
+    'backbone.conv1_(0, 64)_64': un_prune,  # short cut layers
+    'backbone.layer1.0.conv1_(0, 64)_64': stage_ratio_1,
+    'backbone.layer1.1.conv1_(0, 64)_64': stage_ratio_1,
+    'backbone.layer1.2.conv1_(0, 64)_64': un_prune,
+    # stage 2
+    'backbone.layer2.0.conv1_(0, 128)_128': un_prune,
+    'backbone.layer2.0.conv2_(0, 128)_128': un_prune,  # short cut layers
+    'backbone.layer2.1.conv1_(0, 128)_128': stage_ratio_2,
+    'backbone.layer2.2.conv1_(0, 128)_128': stage_ratio_2,
+    'backbone.layer2.3.conv1_(0, 128)_128': un_prune,
+    # stage 3
+    'backbone.layer3.0.conv1_(0, 256)_256': un_prune,
+    'backbone.layer3.0.conv2_(0, 256)_256': un_prune,  # short cut layers
+    'backbone.layer3.1.conv1_(0, 256)_256': stage_ratio_3,
+    'backbone.layer3.2.conv1_(0, 256)_256': stage_ratio_3,
+    'backbone.layer3.3.conv1_(0, 256)_256': stage_ratio_3,
+    'backbone.layer3.4.conv1_(0, 256)_256': stage_ratio_3,
+    'backbone.layer3.5.conv1_(0, 256)_256': un_prune,
+    # stage 4
+    'backbone.layer4.0.conv1_(0, 512)_512': stage_ratio_4,
+    'backbone.layer4.0.conv2_(0, 512)_512': un_prune,  # short cut layers
+    'backbone.layer4.1.conv1_(0, 512)_512': stage_ratio_4,
+    'backbone.layer4.2.conv1_(0, 512)_512': stage_ratio_4
+}
+
+divisor = 8
+##############################################################################
+
+architecture = _base_.model
+
+model = dict(
+    _delete_=True,
+    _scope_='mmrazor',
+    type='GroupFisherDeploySubModel',
+    architecture=architecture,
+    fix_subnet=fix_subnet,
+    divisor=divisor,
+)

configs/pruning/mmcls/l1-norm/l1-norm_resnet34_8xb32_in1k_c_deploy.py

@@ -0,0 +1,54 @@
+#############################################################################
+"""You have to fill these args.
+
+_base_(str): The path to your pretrain config file.
+fix_subnet (Union[dict,str]): The dict store the pruning structure or the
+    json file including it.
+divisor (int): The divisor the make the channel number divisible.
+"""
+
+_base_ = ['mmcls::resnet/resnet34_8xb32_in1k.py']
+un_prune = 1.0
+
+# the config template of target_pruning_ratio can be got by
+# python ./tools/get_channel_units.py {config_file} --choice
+fix_subnet = {
+    # stage 1
+    'backbone.conv1_(0, 64)_64': un_prune,  # short cut layers
+    'backbone.layer1.0.conv1_(0, 64)_64': un_prune,
+    'backbone.layer1.1.conv1_(0, 64)_64': un_prune,
+    'backbone.layer1.2.conv1_(0, 64)_64': un_prune,
+    # stage 2
+    'backbone.layer2.0.conv1_(0, 128)_128': un_prune,
+    'backbone.layer2.0.conv2_(0, 128)_128': un_prune,  # short cut layers
+    'backbone.layer2.1.conv1_(0, 128)_128': un_prune,
+    'backbone.layer2.2.conv1_(0, 128)_128': un_prune,
+    'backbone.layer2.3.conv1_(0, 128)_128': un_prune,
+    # stage 3
+    'backbone.layer3.0.conv1_(0, 256)_256': un_prune,
+    'backbone.layer3.0.conv2_(0, 256)_256': 0.8,  # short cut layers
+    'backbone.layer3.1.conv1_(0, 256)_256': un_prune,
+    'backbone.layer3.2.conv1_(0, 256)_256': un_prune,
+    'backbone.layer3.3.conv1_(0, 256)_256': un_prune,
+    'backbone.layer3.4.conv1_(0, 256)_256': un_prune,
+    'backbone.layer3.5.conv1_(0, 256)_256': un_prune,
+    # stage 4
+    'backbone.layer4.0.conv1_(0, 512)_512': un_prune,
+    'backbone.layer4.0.conv2_(0, 512)_512': un_prune,  # short cut layers
+    'backbone.layer4.1.conv1_(0, 512)_512': un_prune,
+    'backbone.layer4.2.conv1_(0, 512)_512': un_prune
+}
+
+divisor = 8
+##############################################################################
+
+architecture = _base_.model
+
+model = dict(
+    _delete_=True,
+    _scope_='mmrazor',
+    type='GroupFisherDeploySubModel',
+    architecture=architecture,
+    fix_subnet=fix_subnet,
+    divisor=divisor,
+)

configs/pruning/mmcls/l1-norm/metafile.yml

@@ -0,0 +1,28 @@
+Models:
+  - Name: l1-norm_resnet34_8xb32_in1k_a
+    In Collection: L1-norm
+    Config: configs/pruning/mmcls/l1-norm/l1-norm_resnet34_8xb32_in1k_a.py
+    Weights: https://openmmlab-share.oss-cn-hangzhou.aliyuncs.com/mmrazor/v1/pruning/l1-norm/l1-norm_resnet34_8xb32_in1k_a.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 73.61
+  - Name: l1-norm_resnet34_8xb32_in1k_b
+    In Collection: L1-norm
+    Config: configs/pruning/mmcls/l1-norm/l1-norm_resnet34_8xb32_in1k_b.py
+    Weights: https://openmmlab-share.oss-cn-hangzhou.aliyuncs.com/mmrazor/v1/pruning/l1-norm/l1-norm_resnet34_8xb32_in1k_b.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 73.20
+  - Name: l1-norm_resnet34_8xb32_in1k_c
+    In Collection: L1-norm
+    Config: configs/pruning/mmcls/l1-norm/l1-norm_resnet34_8xb32_in1k_c.py
+    Weights: https://openmmlab-share.oss-cn-hangzhou.aliyuncs.com/mmrazor/v1/pruning/l1-norm/l1-norm_resnet34_8xb32_in1k_c.pth
+    Results:
+      - Task: Image Classification
+        Dataset: ImageNet-1k
+        Metrics:
+          Top 1 Accuracy: 73.89

configs/pruning/mmcls/l1-norm/script.sh

@@ -0,0 +1,25 @@
+
+# export pruned checkpoint example
+
+python ./tools/pruning/get_static_model_from_algorithm.py configs/pruning/mmcls/l1-norm/l1-norm_resnet34_8xb32_in1k_a.py https://openmmlab-share.oss-cn-hangzhou.aliyuncs.com/mmrazor/v1/pruning/l1-norm/l1-norm_resnet34_8xb32_in1k_a.pth -o ./work_dirs/norm_resnet34_8xb32_in1k_a
+
+# deploy example
+
+razor_config=configs/pruning/mmcls/l1-norm/l1-norm_resnet34_8xb32_in1k_a_deploy.py
+deploy_config=mmdeploy/configs/mmcls/classification_onnxruntime_dynamic.py
+static_model_checkpoint_path=path/to/pruend/checkpoint
+
+python mmdeploy/tools/deploy.py $deploy_config \
+    $razor_config \
+    $static_model_checkpoint_path \
+    mmdeploy/tests/data/tiger.jpeg \
+    --work-dir ./work_dirs/mmdeploy
+
+python mmdeploy/tools/profiler.py $deploy_config \
+    $razor_config \
+    mmdeploy/demo/resources \
+    --model ./work_dirs/mmdeploy/end2end.onnx \
+    --shape 224x224 \
+    --device cpu \
+    --num-iter 1000 \
+    --warmup 100

configs/pruning/mmdet/group_fisher/retinanet/metafile.yml

@@ -0,0 +1,19 @@
+Models:
+  - Name: group_fisher_act_finetune_retinanet_r50_fpn_1x_coco
+    In Collection: GroupFisher
+    Config: configs/pruning/mmdet/group_fisher/retinanet/group_fisher_act_finetune_retinanet_r50_fpn_1x_coco.py
+    Weights: https://download.openmmlab.com/mmrazor/v1/pruning/group_fisher/retinanet/act/group_fisher_act_finetune_retinanet_r50_fpn_1x_coco.pth
+    Results:
+      - Task: Object Detection
+        Dataset: COCO
+        Metrics:
+          box AP: 36.5
+  - Name: group_fisher_flops_finetune_retinanet_r50_fpn_1x_coco
+    In Collection: GroupFisher
+    Config: configs/pruning/mmdet/group_fisher/retinanet/group_fisher_flops_finetune_retinanet_r50_fpn_1x_coco.py
+    Weights: https://download.openmmlab.com/mmrazor/v1/pruning/group_fisher/retinanet/flops/group_fisher_flops_finetune_retinanet_r50_fpn_1x_coco.pth
+    Results:
+      - Task: Object Detection
+        Dataset: COCO
+        Metrics:
+          box AP: 36.6

model-index.yml

@@ -21,3 +21,8 @@ Import:
   - configs/distill/mmdet/pkd/metafile.yml
   - configs/distill/mmdet3d/pkd/metafile.yml
   - configs/distill/mmcls/deit/metafile.yml
+  - configs/pruning/mmcls/group_fisher/mobilenet/metafile.yml
+  - configs/pruning/mmcls/group_fisher/resnet50/metafile.yml
+  - configs/pruning/mmdet/group_fisher/retinanet/metafile.yml
+  - configs/pruning/mmcls/l1-norm/metafile.yml
+  - configs/pruning/mmcls/dmcp/metafile.yml