# Copyright (c) 2021 Graphcore Ltd. All rights reserved. import argparse opt_to_cfg_map = { "num_ipus": "system.num_ipus", "num_workers": "system.num_workers", "input_channels": "model.input_channels", "activation": "model.activation", "normalization": "model.normalization", "num_classes": "model.n_classes", "class_name_path": "model.class_name_path", "image_size": "model.image_size", "micro_batch_size": "model.micro_batch_size", "mode": "model.mode", "precision": "model.precision", "ipu": "model.ipu", "sharded": "model.sharded", "epochs": "training.epochs", "logging_interval": "training.logging_interval", "exec_cache": "training.exec_cache_path", "checkpoint_interval": "training.checkpoint_interval", "initial_lr": "training.initial_lr", "auto_loss_scaling": "training.auto_loss_scaling", "momentum": "training.momentum", "device_iterations": "ipuopts.device_iterations", "gradient_accumulation": "ipuopts.gradient_accumulation", "class_conf_threshold": "inference.class_conf_threshold", "obj_threshold": "inference.obj_threshold", "iou_threshold": "inference.iou_threshold", "plot_output": "inference.plot_output", "plot_step": "inference.plot_step", "plot_dir": "inference.plot_dir", "plot_threshold": "inference.plot_threshold", "dataset_name": "dataset.name", "max_bbox_per_scale": "dataset.max_bbox_per_scale", "train_file": "dataset.train.file", "test_file": "dataset.test.file", "no_eval": "eval.metrics", "verbose": "eval.verbose", "pre_nms_topk_k": "inference.pre_nms_topk_k", "nms_max_detections": "inference.nms_max_detections", "weight_avg_decay": "training.weight_avg_decay", } def parse_args(): parser = argparse.ArgumentParser() parser.add_argument( "--data", type=str, default="/localdata/datasets/", help="Path to the dataset root dir (default: /localdata/datasets/)", ) parser.add_argument( "--config", type=str, default="configs/inference-yolov4p5.yaml", help="Configuration of the model (default: configs/inference-yolov4p5.yaml)", ) parser.add_argument( "--show-config", action="store_true", default=False, help="Show configuration for the program (default: False)" ) parser.add_argument("--weights", type=str, help="Pretrained weight path to use if specified (default: None)") # The arguments below can be used to overwrite the config parser.add_argument("--num-ipus", type=int, help="Number of IPUs to use (default: 1)") parser.add_argument("--num-workers", type=int, help="Number of workers to use (default: 20)") parser.add_argument("--input-channels", type=int, help="Number of channels in the input image (default: 3)") parser.add_argument("--activation", type=str, help="Activation function to use in the model (default: mish)") parser.add_argument("--normalization", type=str, help="Normalization function to use in the model (default: batch)") parser.add_argument("--num-classes", type=int, help="Number of classes of the model (default: 80)") parser.add_argument("--epochs", type=int, help="Number of training epochs (default: 300)") parser.add_argument( "--class-name-path", type=str, help="Path to the class names yaml (default: ./configs/class_name.yaml)" ) parser.add_argument("--image-size", type=int, help="Size of the input image (default: 896)") parser.add_argument( "--micro-batch-size", type=int, help="The number of samples calculated in one full forward/backward pass (default: 1)", ) parser.add_argument("--mode", type=str, help="Mode to run the model (default: test)") parser.add_argument( "--precision", choices=["half", "mixed", "single"], help="Precision to run the model with (default: half)" ) parser.add_argument( "--benchmark", action="store_true", default=False, help="Run performance benchmark (default: False)" ) parser.add_argument( "--sharded", action="store_true", default=False, help="Use sharded execution, where IPU will sequentially execute a part of the model, \ otherwise pipeline execution will be used instead (default: False)", ) parser.add_argument( "--exec-cache", type=str, help="Path to store training executable cache for the compiled model (default: ./exec_cache)", ) parser.add_argument("--checkpoint-interval", type=int, help="Interval for checkpoint and running validation") parser.add_argument( "--compile-only", action="store_true", help="Create an offline IPU target that can only be used for offline compilation.", ) parser.add_argument("--profile-dir", type=str, help="Enable profiling and store results in this directory") parser.add_argument( "--cpu", dest="ipu", action="store_false", default=True, help="Use cpu to run model (default: True)" ) parser.add_argument("--device-iterations", type=int, help="Number of device iterations (default: 1)") parser.add_argument( "--gradient-accumulation", type=int, help="Number of gradient accumulation, only relevant for training (default: 1", ) parser.add_argument("--initial-lr", type=float, help="Initial learning rate (default: 0.01)") parser.add_argument("--momentum-lr", type=float, help="Initial learning rate (default: 0.937)") parser.add_argument( "--auto-loss-scaling", action="store_true", default=False, help="Use autmatic loss scaling (default is False)" ) parser.add_argument( "--pre-nms-topk-k", type=int, help="Defines how many elements will be chosen to be process in NMS" ) parser.add_argument("--nms-max-detections", type=int, help="Number of output detections after NMS.") parser.add_argument( "--class-conf-threshold", type=float, help="Minimum threshold for class prediction probability (default: 0.4)" ) parser.add_argument("--obj-threshold", type=float, help="Minimum threshold for the objectness score (default: 0.4)") parser.add_argument("--iou-threshold", type=float, help="Minimum threshold for IoU used in NMS (default: 0.65)") parser.add_argument( "--plot-output", action="store_true", default=False, help="Plot the output during inference (default: False)" ) parser.add_argument("--plot-step", type=int, help="Plot every n image (default: 250)") parser.add_argument("--plot-dir", type=str, help="Directory for storing the plot output (default: plots)") parser.add_argument( "--plot-threshold", type=float, help="Minimum threshold for objectness x class prediction for predicted boxes to be plotted (default: 0.3)", ) parser.add_argument("--dataset-name", type=str, help="Name of the dataset (default: coco)") parser.add_argument( "--max-bbox-per-scale", type=int, help="Maximum number of bounding boxes per image (default: 90)" ) parser.add_argument("--train-file", type=str, help="Path to the train annotations (default: train2017.txt)") parser.add_argument("--test-file", type=str, help="Path to the test annotations (default: val2017.txt)") parser.add_argument( "--print-summary", action="store_true", default=False, help="Print out the model architecture (default: False)" ) parser.add_argument( "--no-eval", action="store_false", default=True, help="Dont compute the precision recall metrics (default: True)", ) parser.add_argument( "--verbose", action="store_true", default=False, help="Print out class wise eval (default: False)" ) parser.add_argument( "--wandb", action="store_true", default=False, help="Log metrics to weight and biases (default: False)" ) parser.add_argument( "--logging-interval", type=int, default=200, help="Number of steps to log training progress on console and/or to wandb (default: 200)", ) parser.add_argument( "--weight-avg-decay", type=float, default=0.9999, help="Exponential decay factor for weight averaging" ) return parser.parse_args()