{"owner":"pytorch","repo":"ignite","hasSkills":true,"totalSkillsCount":3,"totalTokensCount":1768,"categories":["anthropic-skill"],"hasMcp":false,"mcpConfig":null,"found":["examples/references/README.md","examples/references/classification/imagenet/README.md","examples/references/segmentation/pascal_voc2012/README.md"],"skills":{"examples/references/README.md":"# Reproducible trainings with Ignite\n\nInspired by [torchvision/references](https://github.com/pytorch/vision/tree/master/references), we provide several\nreproducible baselines for vision tasks:\n\n- [x] Classification\n\n  - [x] [ImageNet](classification/imagenet)\n\n- [x] Segmentation\n  - [x] [Pascal VOC2012](segmentation/pascal_voc2012)\n","examples/references/classification/imagenet/README.md":"# Reproducible ImageNet training with Ignite\n\nIn this example, we provide script and tools to perform reproducible experiments on training neural networks on ImageNet\ndataset.\n\nFeatures:\n\n- Distributed training with native automatic mixed precision\n- Experiments tracking with [ClearML](https://github.com/allegroai/clearml)\n\n| Model     | Training Top-1 Accuracy | Training Top-5 Accuracy | Test Top-1 Accuracy | Test Top-5 Accuracy |\n| --------- | ----------------------- | ----------------------- | ------------------- | ------------------- |\n| ResNet-50 | 78%                     | 92%                     | 77%                 | 94%                 |\n\n| Experiment     | Model | Training Top-1 Accuracy | Training Top-5 Accuracy | Test Top-1 Accuracy | Test Top-5 Accuracy | ClearML Link |\n| -------------- | ----- | ----------------------- | ----------------------- | ------------------- | ------------------- | ------------ |\n| configs/???.py |\n\n## Setup\n\n```\npip install -r requirements.txt\n```\n\n### Docker\n\nFor docker users, you can use the following images to run the example:\n\n```bash\ndocker pull pytorchignite/vision:latest\n```\n\nand install other requirements as suggested above\n\n## Usage\n\nPlease, export the `DATASET_PATH` environment variable for the ImageNet dataset.\n\n```bash\nexport DATASET_PATH=/path/to/imagenet\n# e.g. export DATASET_PATH=/data/ where \"train\", \"val\", \"meta.bin\" are located\n```\n\n### Training\n\n#### Single GPU\n\n- Adjust batch size for your GPU type in the configuration file: `configs/baseline_resnet50.py` or `configs/baseline_resnet50.py`\n\nRun the following command:\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_resnet50.py\n```\n\n#### Multiple GPUs\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_resnet50.py` or `configs/baseline_resnet50.py`\n\n```bash\nOMP_NUM_THREADS=1 torchrun --nproc_per_node=2 main.py training configs/baseline_resnet50.py\n```\n\n## Acknowledgements\n\nTrainings were done using credits provided by [trainml.ai](trainml.ai) platform.\n","examples/references/segmentation/pascal_voc2012/README.md":"# Reproducible PASCAL VOC2012 training with PyTorch-Ignite\n\nIn this example, we provide script and tools to perform reproducible experiments on training neural networks on PASCAL VOC2012\ndataset.\n\nFeatures:\n\n- Distributed training with native automatic mixed precision\n- Experiments tracking with [ClearML](https://github.com/allegroai/clearml)\n\n| Experiment                              | Model               | Dataset  | Val Avg IoU | ClearML Link                                                                                                                         |\n| --------------------------------------- | ------------------- | -------- | ----------- | ------------------------------------------------------------------------------------------------------------------------------------ |\n| configs/baseline_dplv3_resnet101.py     | DeepLabV3 Resnet101 | VOC Only | 0.659161    | [link](https://app.clear.ml/projects/0e9a3a92d3134283b7d5572d516d60c5/experiments/a7254f084a9e47ca9380dfd739f89520/output/execution) |\n| configs/baseline_dplv3_resnet101_sbd.py | DeepLabV3 Resnet101 | VOC+SBD  | 0.6853087   | [link](https://app.clear.ml/projects/0e9a3a92d3134283b7d5572d516d60c5/experiments/dc4cee3377a74d19bc2d0e0e4d638c1f/output/execution) |\n\n## Setup\n\n```\npip install -r requirements.txt\n```\n\n### Docker\n\nFor docker users, you can use the following images to run the example:\n\n```bash\ndocker pull pytorchignite/vision:latest\n```\n\nor\n\n```bash\ndocker pull pytorchignite/hvd-vision:latest\n```\n\nand install other requirements as suggested above\n\n### Using Horovod as distributed framework\n\nWe do not add `horovod` as a requirement into `requirements.txt`. Please, install it manually following the official guides or\nuse `pytorchignite/hvd-vision:latest` docker image.\n\n### (Optional) Download Pascal VOC2012 and SDB datasets\n\nDownload and extract the datasets:\n\n```bash\npython main.py download /path/to/datasets\n```\n\nThis script will download and extract the following datasets into `/path/to/datasets`\n\n- The [Pascal VOC2012](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar) dataset\n- Optionally, the [SBD](http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/semantic_contours/benchmark.tgz) evaluation dataset\n\n## Usage\n\nPlease, export the `DATASET_PATH` environment variable for the Pascal VOC2012 dataset.\n\n```bash\nexport DATASET_PATH=/path/to/pascal_voc2012\n# e.g. export DATASET_PATH=/data/ where VOCdevkit is located\n```\n\nOptionally, if using SBD dataset, export the `SBD_DATASET_PATH` environment variable:\n\n```bash\nexport SBD_DATASET_PATH=/path/to/SBD/\n# e.g. export SBD_DATASET_PATH=/data/SBD/  where \"cls  img  inst  train.txt  train_noval.txt  val.txt\" are located\n```\n\n### Training\n\n#### Single GPU\n\n- Adjust batch size for your GPU type in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\nRun the following command:\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_dplv3_resnet101_sbd.py\n# or without SBD\n# CUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_dplv3_resnet101.py\n```\n\n#### Multiple GPUs\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\n```bash\ntorchrun --nproc_per_node=2 main.py training configs/baseline_dplv3_resnet101_sbd.py\n# or without SBD\n# torchrun --nproc_per_node=2 main.py training configs/baseline_dplv3_resnet101.py\n```\n\n#### Using Horovod as distributed framework\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\n```bash\nhorovodrun -np=2 python -u main.py training configs/baseline_dplv3_resnet101_sbd.py --backend=\"horovod\"\n# or without SBD\n# horovodrun -np=2 python -u main.py training configs/baseline_dplv3_resnet101.py --backend=\"horovod\"\n```\n\n### Evaluation\n\n#### Single GPU\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py\n```\n\n#### Multiple GPUs\n\n```bash\ntorchrun --nproc_per_node=2 main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py\n```\n\n#### Using Horovod as distributed framework\n\n```bash\nhorovodrun -np=2 python -u main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py --backend=\"horovod\"\n```\n\n## Acknowledgements\n\nTrainings were done using credits provided by AWS for open-source development via NumFOCUS\nand using [trainml.ai](trainml.ai) platform.\n"},"files":{"examples/references/README.md":"# Reproducible trainings with Ignite\n\nInspired by [torchvision/references](https://github.com/pytorch/vision/tree/master/references), we provide several\nreproducible baselines for vision tasks:\n\n- [x] Classification\n\n  - [x] [ImageNet](classification/imagenet)\n\n- [x] Segmentation\n  - [x] [Pascal VOC2012](segmentation/pascal_voc2012)\n","examples/references/classification/imagenet/README.md":"# Reproducible ImageNet training with Ignite\n\nIn this example, we provide script and tools to perform reproducible experiments on training neural networks on ImageNet\ndataset.\n\nFeatures:\n\n- Distributed training with native automatic mixed precision\n- Experiments tracking with [ClearML](https://github.com/allegroai/clearml)\n\n| Model     | Training Top-1 Accuracy | Training Top-5 Accuracy | Test Top-1 Accuracy | Test Top-5 Accuracy |\n| --------- | ----------------------- | ----------------------- | ------------------- | ------------------- |\n| ResNet-50 | 78%                     | 92%                     | 77%                 | 94%                 |\n\n| Experiment     | Model | Training Top-1 Accuracy | Training Top-5 Accuracy | Test Top-1 Accuracy | Test Top-5 Accuracy | ClearML Link |\n| -------------- | ----- | ----------------------- | ----------------------- | ------------------- | ------------------- | ------------ |\n| configs/???.py |\n\n## Setup\n\n```\npip install -r requirements.txt\n```\n\n### Docker\n\nFor docker users, you can use the following images to run the example:\n\n```bash\ndocker pull pytorchignite/vision:latest\n```\n\nand install other requirements as suggested above\n\n## Usage\n\nPlease, export the `DATASET_PATH` environment variable for the ImageNet dataset.\n\n```bash\nexport DATASET_PATH=/path/to/imagenet\n# e.g. export DATASET_PATH=/data/ where \"train\", \"val\", \"meta.bin\" are located\n```\n\n### Training\n\n#### Single GPU\n\n- Adjust batch size for your GPU type in the configuration file: `configs/baseline_resnet50.py` or `configs/baseline_resnet50.py`\n\nRun the following command:\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_resnet50.py\n```\n\n#### Multiple GPUs\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_resnet50.py` or `configs/baseline_resnet50.py`\n\n```bash\nOMP_NUM_THREADS=1 torchrun --nproc_per_node=2 main.py training configs/baseline_resnet50.py\n```\n\n## Acknowledgements\n\nTrainings were done using credits provided by [trainml.ai](trainml.ai) platform.\n","examples/references/segmentation/pascal_voc2012/README.md":"# Reproducible PASCAL VOC2012 training with PyTorch-Ignite\n\nIn this example, we provide script and tools to perform reproducible experiments on training neural networks on PASCAL VOC2012\ndataset.\n\nFeatures:\n\n- Distributed training with native automatic mixed precision\n- Experiments tracking with [ClearML](https://github.com/allegroai/clearml)\n\n| Experiment                              | Model               | Dataset  | Val Avg IoU | ClearML Link                                                                                                                         |\n| --------------------------------------- | ------------------- | -------- | ----------- | ------------------------------------------------------------------------------------------------------------------------------------ |\n| configs/baseline_dplv3_resnet101.py     | DeepLabV3 Resnet101 | VOC Only | 0.659161    | [link](https://app.clear.ml/projects/0e9a3a92d3134283b7d5572d516d60c5/experiments/a7254f084a9e47ca9380dfd739f89520/output/execution) |\n| configs/baseline_dplv3_resnet101_sbd.py | DeepLabV3 Resnet101 | VOC+SBD  | 0.6853087   | [link](https://app.clear.ml/projects/0e9a3a92d3134283b7d5572d516d60c5/experiments/dc4cee3377a74d19bc2d0e0e4d638c1f/output/execution) |\n\n## Setup\n\n```\npip install -r requirements.txt\n```\n\n### Docker\n\nFor docker users, you can use the following images to run the example:\n\n```bash\ndocker pull pytorchignite/vision:latest\n```\n\nor\n\n```bash\ndocker pull pytorchignite/hvd-vision:latest\n```\n\nand install other requirements as suggested above\n\n### Using Horovod as distributed framework\n\nWe do not add `horovod` as a requirement into `requirements.txt`. Please, install it manually following the official guides or\nuse `pytorchignite/hvd-vision:latest` docker image.\n\n### (Optional) Download Pascal VOC2012 and SDB datasets\n\nDownload and extract the datasets:\n\n```bash\npython main.py download /path/to/datasets\n```\n\nThis script will download and extract the following datasets into `/path/to/datasets`\n\n- The [Pascal VOC2012](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar) dataset\n- Optionally, the [SBD](http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/semantic_contours/benchmark.tgz) evaluation dataset\n\n## Usage\n\nPlease, export the `DATASET_PATH` environment variable for the Pascal VOC2012 dataset.\n\n```bash\nexport DATASET_PATH=/path/to/pascal_voc2012\n# e.g. export DATASET_PATH=/data/ where VOCdevkit is located\n```\n\nOptionally, if using SBD dataset, export the `SBD_DATASET_PATH` environment variable:\n\n```bash\nexport SBD_DATASET_PATH=/path/to/SBD/\n# e.g. export SBD_DATASET_PATH=/data/SBD/  where \"cls  img  inst  train.txt  train_noval.txt  val.txt\" are located\n```\n\n### Training\n\n#### Single GPU\n\n- Adjust batch size for your GPU type in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\nRun the following command:\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_dplv3_resnet101_sbd.py\n# or without SBD\n# CUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_dplv3_resnet101.py\n```\n\n#### Multiple GPUs\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\n```bash\ntorchrun --nproc_per_node=2 main.py training configs/baseline_dplv3_resnet101_sbd.py\n# or without SBD\n# torchrun --nproc_per_node=2 main.py training configs/baseline_dplv3_resnet101.py\n```\n\n#### Using Horovod as distributed framework\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\n```bash\nhorovodrun -np=2 python -u main.py training configs/baseline_dplv3_resnet101_sbd.py --backend=\"horovod\"\n# or without SBD\n# horovodrun -np=2 python -u main.py training configs/baseline_dplv3_resnet101.py --backend=\"horovod\"\n```\n\n### Evaluation\n\n#### Single GPU\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py\n```\n\n#### Multiple GPUs\n\n```bash\ntorchrun --nproc_per_node=2 main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py\n```\n\n#### Using Horovod as distributed framework\n\n```bash\nhorovodrun -np=2 python -u main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py --backend=\"horovod\"\n```\n\n## Acknowledgements\n\nTrainings were done using credits provided by AWS for open-source development via NumFOCUS\nand using [trainml.ai](trainml.ai) platform.\n"},"items":[{"name":"README.md","path":"examples/references/classification/imagenet/README.md","rawUrl":"https://raw.githubusercontent.com/pytorch/ignite/HEAD/examples/references/classification/imagenet/README.md","title":"Imagenet Skill","category":"anthropic-skill","format":"markdown","content":"# Reproducible ImageNet training with Ignite\n\nIn this example, we provide script and tools to perform reproducible experiments on training neural networks on ImageNet\ndataset.\n\nFeatures:\n\n- Distributed training with native automatic mixed precision\n- Experiments tracking with [ClearML](https://github.com/allegroai/clearml)\n\n| Model     | Training Top-1 Accuracy | Training Top-5 Accuracy | Test Top-1 Accuracy | Test Top-5 Accuracy |\n| --------- | ----------------------- | ----------------------- | ------------------- | ------------------- |\n| ResNet-50 | 78%                     | 92%                     | 77%                 | 94%                 |\n\n| Experiment     | Model | Training Top-1 Accuracy | Training Top-5 Accuracy | Test Top-1 Accuracy | Test Top-5 Accuracy | ClearML Link |\n| -------------- | ----- | ----------------------- | ----------------------- | ------------------- | ------------------- | ------------ |\n| configs/???.py |\n\n## Setup\n\n```\npip install -r requirements.txt\n```\n\n### Docker\n\nFor docker users, you can use the following images to run the example:\n\n```bash\ndocker pull pytorchignite/vision:latest\n```\n\nand install other requirements as suggested above\n\n## Usage\n\nPlease, export the `DATASET_PATH` environment variable for the ImageNet dataset.\n\n```bash\nexport DATASET_PATH=/path/to/imagenet\n# e.g. export DATASET_PATH=/data/ where \"train\", \"val\", \"meta.bin\" are located\n```\n\n### Training\n\n#### Single GPU\n\n- Adjust batch size for your GPU type in the configuration file: `configs/baseline_resnet50.py` or `configs/baseline_resnet50.py`\n\nRun the following command:\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_resnet50.py\n```\n\n#### Multiple GPUs\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_resnet50.py` or `configs/baseline_resnet50.py`\n\n```bash\nOMP_NUM_THREADS=1 torchrun --nproc_per_node=2 main.py training configs/baseline_resnet50.py\n```\n\n## Acknowledgements\n\nTrainings were done using credits provided by [trainml.ai](trainml.ai) platform.\n","isInternal":false,"tokens":472,"sizeBytes":2056},{"name":"README.md","path":"examples/references/README.md","rawUrl":"https://raw.githubusercontent.com/pytorch/ignite/HEAD/examples/references/README.md","title":"References Skill","category":"anthropic-skill","format":"markdown","content":"# Reproducible trainings with Ignite\n\nInspired by [torchvision/references](https://github.com/pytorch/vision/tree/master/references), we provide several\nreproducible baselines for vision tasks:\n\n- [x] Classification\n\n  - [x] [ImageNet](classification/imagenet)\n\n- [x] Segmentation\n  - [x] [Pascal VOC2012](segmentation/pascal_voc2012)\n","isInternal":false,"tokens":94,"sizeBytes":335},{"name":"README.md","path":"examples/references/segmentation/pascal_voc2012/README.md","rawUrl":"https://raw.githubusercontent.com/pytorch/ignite/HEAD/examples/references/segmentation/pascal_voc2012/README.md","title":"Pascal_voc2012 Skill","category":"anthropic-skill","format":"markdown","content":"# Reproducible PASCAL VOC2012 training with PyTorch-Ignite\n\nIn this example, we provide script and tools to perform reproducible experiments on training neural networks on PASCAL VOC2012\ndataset.\n\nFeatures:\n\n- Distributed training with native automatic mixed precision\n- Experiments tracking with [ClearML](https://github.com/allegroai/clearml)\n\n| Experiment                              | Model               | Dataset  | Val Avg IoU | ClearML Link                                                                                                                         |\n| --------------------------------------- | ------------------- | -------- | ----------- | ------------------------------------------------------------------------------------------------------------------------------------ |\n| configs/baseline_dplv3_resnet101.py     | DeepLabV3 Resnet101 | VOC Only | 0.659161    | [link](https://app.clear.ml/projects/0e9a3a92d3134283b7d5572d516d60c5/experiments/a7254f084a9e47ca9380dfd739f89520/output/execution) |\n| configs/baseline_dplv3_resnet101_sbd.py | DeepLabV3 Resnet101 | VOC+SBD  | 0.6853087   | [link](https://app.clear.ml/projects/0e9a3a92d3134283b7d5572d516d60c5/experiments/dc4cee3377a74d19bc2d0e0e4d638c1f/output/execution) |\n\n## Setup\n\n```\npip install -r requirements.txt\n```\n\n### Docker\n\nFor docker users, you can use the following images to run the example:\n\n```bash\ndocker pull pytorchignite/vision:latest\n```\n\nor\n\n```bash\ndocker pull pytorchignite/hvd-vision:latest\n```\n\nand install other requirements as suggested above\n\n### Using Horovod as distributed framework\n\nWe do not add `horovod` as a requirement into `requirements.txt`. Please, install it manually following the official guides or\nuse `pytorchignite/hvd-vision:latest` docker image.\n\n### (Optional) Download Pascal VOC2012 and SDB datasets\n\nDownload and extract the datasets:\n\n```bash\npython main.py download /path/to/datasets\n```\n\nThis script will download and extract the following datasets into `/path/to/datasets`\n\n- The [Pascal VOC2012](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar) dataset\n- Optionally, the [SBD](http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/semantic_contours/benchmark.tgz) evaluation dataset\n\n## Usage\n\nPlease, export the `DATASET_PATH` environment variable for the Pascal VOC2012 dataset.\n\n```bash\nexport DATASET_PATH=/path/to/pascal_voc2012\n# e.g. export DATASET_PATH=/data/ where VOCdevkit is located\n```\n\nOptionally, if using SBD dataset, export the `SBD_DATASET_PATH` environment variable:\n\n```bash\nexport SBD_DATASET_PATH=/path/to/SBD/\n# e.g. export SBD_DATASET_PATH=/data/SBD/  where \"cls  img  inst  train.txt  train_noval.txt  val.txt\" are located\n```\n\n### Training\n\n#### Single GPU\n\n- Adjust batch size for your GPU type in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\nRun the following command:\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_dplv3_resnet101_sbd.py\n# or without SBD\n# CUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_dplv3_resnet101.py\n```\n\n#### Multiple GPUs\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\n```bash\ntorchrun --nproc_per_node=2 main.py training configs/baseline_dplv3_resnet101_sbd.py\n# or without SBD\n# torchrun --nproc_per_node=2 main.py training configs/baseline_dplv3_resnet101.py\n```\n\n#### Using Horovod as distributed framework\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\n```bash\nhorovodrun -np=2 python -u main.py training configs/baseline_dplv3_resnet101_sbd.py --backend=\"horovod\"\n# or without SBD\n# horovodrun -np=2 python -u main.py training configs/baseline_dplv3_resnet101.py --backend=\"horovod\"\n```\n\n### Evaluation\n\n#### Single GPU\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py\n```\n\n#### Multiple GPUs\n\n```bash\ntorchrun --nproc_per_node=2 main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py\n```\n\n#### Using Horovod as distributed framework\n\n```bash\nhorovodrun -np=2 python -u main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py --backend=\"horovod\"\n```\n\n## Acknowledgements\n\nTrainings were done using credits provided by AWS for open-source development via NumFOCUS\nand using [trainml.ai](trainml.ai) platform.\n","isInternal":false,"tokens":1202,"sizeBytes":4522}],"systemPromptSnippet":"<yakaai_skills repo=\"pytorch/ignite\">\n<!-- File: examples/references/classification/imagenet/README.md (Tokens: ~472 | Category: anthropic-skill) -->\n# Reproducible ImageNet training with Ignite\n\nIn this example, we provide script and tools to perform reproducible experiments on training neural networks on ImageNet\ndataset.\n\nFeatures:\n\n- Distributed training with native automatic mixed precision\n- Experiments tracking with [ClearML](https://github.com/allegroai/clearml)\n\n| Model     | Training Top-1 Accuracy | Training Top-5 Accuracy | Test Top-1 Accuracy | Test Top-5 Accuracy |\n| --------- | ----------------------- | ----------------------- | ------------------- | ------------------- |\n| ResNet-50 | 78%                     | 92%                     | 77%                 | 94%                 |\n\n| Experiment     | Model | Training Top-1 Accuracy | Training Top-5 Accuracy | Test Top-1 Accuracy | Test Top-5 Accuracy | ClearML Link |\n| -------------- | ----- | ----------------------- | ----------------------- | ------------------- | ------------------- | ------------ |\n| configs/???.py |\n\n## Setup\n\n```\npip install -r requirements.txt\n```\n\n### Docker\n\nFor docker users, you can use the following images to run the example:\n\n```bash\ndocker pull pytorchignite/vision:latest\n```\n\nand install other requirements as suggested above\n\n## Usage\n\nPlease, export the `DATASET_PATH` environment variable for the ImageNet dataset.\n\n```bash\nexport DATASET_PATH=/path/to/imagenet\n# e.g. export DATASET_PATH=/data/ where \"train\", \"val\", \"meta.bin\" are located\n```\n\n### Training\n\n#### Single GPU\n\n- Adjust batch size for your GPU type in the configuration file: `configs/baseline_resnet50.py` or `configs/baseline_resnet50.py`\n\nRun the following command:\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_resnet50.py\n```\n\n#### Multiple GPUs\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_resnet50.py` or `configs/baseline_resnet50.py`\n\n```bash\nOMP_NUM_THREADS=1 torchrun --nproc_per_node=2 main.py training configs/baseline_resnet50.py\n```\n\n## Acknowledgements\n\nTrainings were done using credits provided by [trainml.ai](trainml.ai) platform.\n\n<!-- File: examples/references/README.md (Tokens: ~94 | Category: anthropic-skill) -->\n# Reproducible trainings with Ignite\n\nInspired by [torchvision/references](https://github.com/pytorch/vision/tree/master/references), we provide several\nreproducible baselines for vision tasks:\n\n- [x] Classification\n\n  - [x] [ImageNet](classification/imagenet)\n\n- [x] Segmentation\n  - [x] [Pascal VOC2012](segmentation/pascal_voc2012)\n\n<!-- File: examples/references/segmentation/pascal_voc2012/README.md (Tokens: ~1202 | Category: anthropic-skill) -->\n# Reproducible PASCAL VOC2012 training with PyTorch-Ignite\n\nIn this example, we provide script and tools to perform reproducible experiments on training neural networks on PASCAL VOC2012\ndataset.\n\nFeatures:\n\n- Distributed training with native automatic mixed precision\n- Experiments tracking with [ClearML](https://github.com/allegroai/clearml)\n\n| Experiment                              | Model               | Dataset  | Val Avg IoU | ClearML Link                                                                                                                         |\n| --------------------------------------- | ------------------- | -------- | ----------- | ------------------------------------------------------------------------------------------------------------------------------------ |\n| configs/baseline_dplv3_resnet101.py     | DeepLabV3 Resnet101 | VOC Only | 0.659161    | [link](https://app.clear.ml/projects/0e9a3a92d3134283b7d5572d516d60c5/experiments/a7254f084a9e47ca9380dfd739f89520/output/execution) |\n| configs/baseline_dplv3_resnet101_sbd.py | DeepLabV3 Resnet101 | VOC+SBD  | 0.6853087   | [link](https://app.clear.ml/projects/0e9a3a92d3134283b7d5572d516d60c5/experiments/dc4cee3377a74d19bc2d0e0e4d638c1f/output/execution) |\n\n## Setup\n\n```\npip install -r requirements.txt\n```\n\n### Docker\n\nFor docker users, you can use the following images to run the example:\n\n```bash\ndocker pull pytorchignite/vision:latest\n```\n\nor\n\n```bash\ndocker pull pytorchignite/hvd-vision:latest\n```\n\nand install other requirements as suggested above\n\n### Using Horovod as distributed framework\n\nWe do not add `horovod` as a requirement into `requirements.txt`. Please, install it manually following the official guides or\nuse `pytorchignite/hvd-vision:latest` docker image.\n\n### (Optional) Download Pascal VOC2012 and SDB datasets\n\nDownload and extract the datasets:\n\n```bash\npython main.py download /path/to/datasets\n```\n\nThis script will download and extract the following datasets into `/path/to/datasets`\n\n- The [Pascal VOC2012](http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar) dataset\n- Optionally, the [SBD](http://www.eecs.berkeley.edu/Research/Projects/CS/vision/grouping/semantic_contours/benchmark.tgz) evaluation dataset\n\n## Usage\n\nPlease, export the `DATASET_PATH` environment variable for the Pascal VOC2012 dataset.\n\n```bash\nexport DATASET_PATH=/path/to/pascal_voc2012\n# e.g. export DATASET_PATH=/data/ where VOCdevkit is located\n```\n\nOptionally, if using SBD dataset, export the `SBD_DATASET_PATH` environment variable:\n\n```bash\nexport SBD_DATASET_PATH=/path/to/SBD/\n# e.g. export SBD_DATASET_PATH=/data/SBD/  where \"cls  img  inst  train.txt  train_noval.txt  val.txt\" are located\n```\n\n### Training\n\n#### Single GPU\n\n- Adjust batch size for your GPU type in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\nRun the following command:\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_dplv3_resnet101_sbd.py\n# or without SBD\n# CUDA_VISIBLE_DEVICES=0 python -u main.py training configs/baseline_dplv3_resnet101.py\n```\n\n#### Multiple GPUs\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\n```bash\ntorchrun --nproc_per_node=2 main.py training configs/baseline_dplv3_resnet101_sbd.py\n# or without SBD\n# torchrun --nproc_per_node=2 main.py training configs/baseline_dplv3_resnet101.py\n```\n\n#### Using Horovod as distributed framework\n\n- Adjust total batch size for your GPUs in the configuration file: `configs/baseline_dplv3_resnet101_sbd.py` or `configs/baseline_dplv3_resnet101.py`\n\n```bash\nhorovodrun -np=2 python -u main.py training configs/baseline_dplv3_resnet101_sbd.py --backend=\"horovod\"\n# or without SBD\n# horovodrun -np=2 python -u main.py training configs/baseline_dplv3_resnet101.py --backend=\"horovod\"\n```\n\n### Evaluation\n\n#### Single GPU\n\n```bash\nCUDA_VISIBLE_DEVICES=0 python -u main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py\n```\n\n#### Multiple GPUs\n\n```bash\ntorchrun --nproc_per_node=2 main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py\n```\n\n#### Using Horovod as distributed framework\n\n```bash\nhorovodrun -np=2 python -u main.py eval configs/eval_baseline_dplv3_resnet101_sbd.py --backend=\"horovod\"\n```\n\n## Acknowledgements\n\nTrainings were done using credits provided by AWS for open-source development via NumFOCUS\nand using [trainml.ai](trainml.ai) platform.\n</yakaai_skills>"}