batch_train_cityscape_benchmark.py

import os
import subprocess
from itertools import product

# v_list = ['dashcam_%d_test' % (i+1) for i in range(4)] + ['trafficcam_%d_test' % (i+1) for i in range(4)]
# v_list = [v_list[0]]

# v_list = ['train_first/trafficcam_%d_train' % (i+1) for i in range(4)] + ['train_first/dashcam_%d_train' % (i+1) for i in range(4)]
# v_list = [v_list[4]]
app = "COCO-Detection/faster_rcnn_R_101_FPN_3x.yaml"
# app = "EfficientDet"
# model_name = f"COCO_full_normalizedsaliency_R_101_FPN_crossthresh_5xdownsample"
architecture = "SSD"

# model_name = "visdrone_R_101_FPN_crossthresh"
filename = "benchmark_fcn"
gt = "pickles/COCO_saliency_EfficientDet_withtestset_withoutclasscheck.pickle"

for compute in [0.5]:

    weight = 4

    model_name = f"compute_{compute}_weight_{weight}"

    print(f"Compute: {compute}, Weight: {weight}")

    subprocess.run(
        [
            "python",
            "train_cityscape_benchmark.py",
            "--training_set",
            "COCO",
            "--no_class_check",
            "-g",
            # f"visdrone_normalizedsaliency_R_101_FPN.pickle",
            f"{gt}",
            "-p",
            f"maskgen_pths/{model_name}.pth",
            # "--init",
            # f"maskgen_pths/{model_name}.pth.best",
            "--tile_size",
            "16",
            "--batch_size",
            "4",
            "--log",
            f"train_{model_name}.log",
            "--maskgen_file",
            f"/tank/kuntai/code/video-compression/maskgen/{filename}.py",
            "--visualize",
            "True",
            "--visualize_step_size",
            "1000000",
            "--app",
            # f"Segmentation/fcn_resnet50",
            f"{app}",
            "--local_rank",
            "-1",
            "--num_workers",
            "10",
            "--learning_rate",
            "1e-3",
            "--test_set",
            "object_detection_test_set",
            "--confidence_threshold",
            "0.3",
            "--gt_confidence_threshold",
            "0.3",
            "--weight",
            f"{weight}",
            "--compute",
            f"{compute}",
        ]
    )