gen_eval_table1.py

from itertools import product
import sys
import os
import json

rootdir = os.path.abspath(sys.path[0] + '/../')

datadir = f"{rootdir}/data/__new/classroom_fovea_r360x80_t0.6"
n_nets_arr = [ 1, 2, 4, 8 ]
nf_arr = [ 64, 128, 256, 512, 1024 ]
n_layers_arr = [ 2, 4, 8 ]

head = "Nets," + ",".join([f"{val1}x{val2}" for val1, val2 in product(nf_arr, n_layers_arr)])
perf_train_table = []
perf_test_table = []
perf_time_table = []
for n_nets in n_nets_arr:
    perf_train_row = []
    perf_test_row = []
    perf_time_row = []
    for nf, n_layers in product(nf_arr, n_layers_arr):
        
        configid = f"eval@snerffast{n_nets}-rgb_e6_fc{nf}x{n_layers}_d1.00-7.00_s64_~p"
        outputdir = f"{datadir}/{configid}/output_50"
        if not os.path.exists(outputdir):
            perf_train_row.append("-")
            perf_test_row.append("-")
            perf_time_row.append("-")
            continue
        perf_test_found=False
        perf_train_found=False
        for file in os.listdir(outputdir):
            if file.startswith("perf_r120x80_test"):
                if perf_test_found:
                    os.remove(f"{outputdir}/{file}")
                else:
                    perf_test_row.append(os.path.splitext(file)[0].split("_")[-1])
                    perf_test_found=True
            elif file.startswith("perf_r120x80"):
                if perf_train_found:
                    os.remove(f"{outputdir}/{file}")
                else:
                    perf_train_row.append(os.path.splitext(file)[0].split("_")[-1])
                    perf_train_found=True
        if perf_train_found == False:
            perf_train_row.append("-")
        if perf_test_found == False:
            perf_test_row.append("-")
        # Collect time values
        time_file = f"{datadir}/eval_trt/time/eval_{n_nets}x{nf}x{n_layers}.json"
        if not os.path.exists(time_file):
            perf_time_row.append("-")
        else:
            with open(time_file) as fp:
                time_data = json.load(fp)
            time = 0
            for item in time_data:
                time += item['computeMs']
            time /= len(time_data)
            perf_time_row.append(f"{time:.1f}")

    perf_train_table.append(perf_train_row)
    perf_test_table.append(perf_test_row)
    perf_time_table.append(perf_time_row)

perf_train_content = head + "\n"
for i, row in enumerate(perf_train_table):
    perf_train_content += f"{n_nets_arr[i]},"
    perf_train_content += ",".join(row) + "\n"

perf_test_content = head + "\n"
for i, row in enumerate(perf_test_table):
    perf_test_content += f"{n_nets_arr[i]},"
    perf_test_content += ",".join(row) + "\n"

perf_time_content = head + "\n"
for i, row in enumerate(perf_time_table):
    perf_time_content += f"{n_nets_arr[i]},"
    perf_time_content += ",".join(row) + "\n"

with open(f"{datadir}/eval_perf.csv", "w") as fp:
    fp.write("Train:\n")
    fp.write(perf_train_content)
    fp.write("Test:\n")
    fp.write(perf_test_content)
    fp.write("Time:\n")
    fp.write(perf_time_content)