Delete scripts

scripts/.ipynb_checkpoints/test-checkpoint.sh

@@ -1,40 +0,0 @@
-# #expid: 1.a
-# python -W ignore::UserWarning tools/test.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0,1,2]" \
-#     TEST.RESUME 1.a-v1e3_n1e2_012 \
-#     TEST.DATASET "unseen" \
-#     TEST.MODEL "seen" \
-#     TEST.EPOCH 85
-    
-# #expid: 1.b
-# python -W ignore::UserWarning tools/test.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     TEST.RESUME 1.b-v1e3_n1e2_0 \
-#     TEST.DATASET "unseen" \
-#     TEST.MODEL "seen" \
-#     TEST.EPOCH 63
-    
-
-# #expid: 1.c
-# python -W ignore::UserWarning tools/test.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[1]" \
-#     TEST.RESUME 1.c-v1e3_n1e2_1 \
-#     TEST.DATASET "unseen" \
-#     TEST.MODEL "seen" \
-#     TEST.EPOCH 65
-
-#expid: 1.d
-python -W ignore::UserWarning tools/test.py --cfg configs/selfattention_noise.yaml \
-    MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-    MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-    MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[2]" \
-    TEST.RESUME 1.d-v1e3_n1e2_2 \
-    TEST.DATASET "unseen" \
-    TEST.MODEL "seen" \
-    TEST.EPOCH 41

scripts/.ipynb_checkpoints/train-checkpoint.sh

@@ -1,117 +0,0 @@
-# #expid: 0.a
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention.yaml \
-#     SESSION 0.a
-    
-# #expid: 0.b
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention.yaml \
-#     SESSION 0.b
-
-# #expid: 0.c # change from utils deterministic to original method in coattention
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention.yaml \
-#     SESSION 0.c
-
-# #expid: 0.d # use deterministic 123 as coattention
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention.yaml \
-#     SESSION 0.d
-
-
-# 1 compare adding noise at different layers
-# #expid: 1.a
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0,1,2]" \
-#     SESSION 1.a-v1e3_n1e2_012
-    
-# #expid: 1.b
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     SESSION 1.b-v1e3_n1e2_0
-    
-# #expid: 1.c
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[1]" \
-#     SESSION 1.c-v1e3_n1e2_1
-    
-# #expid: 1.d
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[2]" \
-#     SESSION 1.d-v1e3_n1e2_2
-    
-    
-# #expid: 2.a
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     TRAIN.SAMPLE 1    \
-#     SESSION 2.a-v1e3_n1e2_0_0.5
-    
-# #expid: 3.a
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     SESSION 3.a-v1e2_n1e2_0
-    
-# #expid: 3.b
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-4 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     SESSION 3.b-v1e4_n1e2_0
-    
-# #expid: 3.c
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-5 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     SESSION 3.c-v1e5_n1e2_0
-  
-  
-# # try different noise type
-# #expid: 4.a
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     MODEL.SELFATTENTION.NOISE_TYPE "blurry" \
-#     SESSION 4.a_v1e3_n1e2_0_blurry
-
-
-# #expid: 4.b  # to be run
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     MODEL.SELFATTENTION.NOISE_TYPE "adaptive" \
-#     SESSION 4.b_v1e3_n1e2_0_adaptive
-
-    
-# #expid: 5.a
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     SESSION 5.a-v1e3_n1e3_0
-    
-# #expid: 5.b  # zekrom
-# python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-#     MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-3 \
-#     MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-4 \
-#     MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[0]" \
-#     SESSION 5.b-v1e3_n1e4_0
-
-#expid: 6.a
-python -W ignore::UserWarning tools/train.py --cfg configs/selfattention_noise.yaml \
-    MODEL.SELFATTENTION.VERB_BASE_NOISE 1e-4 \
-    MODEL.SELFATTENTION.NOUN_BASE_NOISE 1e-2 \
-    MODEL.SELFATTENTION.ADD_NOISE_LAYERS "[1]" \
-    SESSION 6.a-v1e4_n1e2_1
-    

scripts/.ipynb_checkpoints/try-checkpoint.ipynb

@@ -1,154 +0,0 @@
-{
- "cells": [
-  {
-   "cell_type": "code",
-   "execution_count": 1,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "import math\n",
-    "import numbers\n",
-    "import torch\n",
-    "from torch import nn\n",
-    "from torch.nn import functional as F\n",
-    "\n",
-    "class GaussianSmoothing(nn.Module):\n",
-    "    \"\"\"\n",
-    "    Apply gaussian smoothing on a\n",
-    "    1d, 2d or 3d tensor. Filtering is performed seperately for each channel\n",
-    "    in the input using a depthwise convolution.\n",
-    "    Arguments:\n",
-    "        channels (int, sequence): Number of channels of the input tensors. Output will\n",
-    "            have this number of channels as well.\n",
-    "        kernel_size (int, sequence): Size of the gaussian kernel.\n",
-    "        sigma (float, sequence): Standard deviation of the gaussian kernel.\n",
-    "        dim (int, optional): The number of dimensions of the data.\n",
-    "            Default value is 2 (spatial).\n",
-    "    \"\"\"\n",
-    "    def __init__(self, channels, kernel_size, sigma, dim=2):\n",
-    "        super(GaussianSmoothing, self).__init__()\n",
-    "        if isinstance(kernel_size, numbers.Number):\n",
-    "            kernel_size = [kernel_size] * dim\n",
-    "        if isinstance(sigma, numbers.Number):\n",
-    "            sigma = [sigma] * dim\n",
-    "\n",
-    "        # The gaussian kernel is the product of the\n",
-    "        # gaussian function of each dimension.\n",
-    "        kernel = 1\n",
-    "        meshgrids = torch.meshgrid(\n",
-    "            [\n",
-    "                torch.arange(size, dtype=torch.float32)\n",
-    "                for size in kernel_size\n",
-    "            ]\n",
-    "        )\n",
-    "        for size, std, mgrid in zip(kernel_size, sigma, meshgrids):\n",
-    "            mean = (size - 1) / 2\n",
-    "            kernel *= 1 / (std * math.sqrt(2 * math.pi)) * \\\n",
-    "                      torch.exp(-((mgrid - mean) / std) ** 2 / 2)\n",
-    "\n",
-    "        # Make sure sum of values in gaussian kernel equals 1.\n",
-    "        kernel = kernel / torch.sum(kernel)\n",
-    "\n",
-    "        # Reshape to depthwise convolutional weight\n",
-    "        kernel = kernel.view(1, 1, *kernel.size())\n",
-    "        kernel = kernel.repeat(channels, *[1] * (kernel.dim() - 1))\n",
-    "\n",
-    "        self.register_buffer('weight', kernel)\n",
-    "        self.groups = channels\n",
-    "\n",
-    "        if dim == 1:\n",
-    "            self.conv = F.conv1d\n",
-    "        elif dim == 2:\n",
-    "            self.conv = F.conv2d\n",
-    "        elif dim == 3:\n",
-    "            self.conv = F.conv3d\n",
-    "        else:\n",
-    "            raise RuntimeError(\n",
-    "                'Only 1, 2 and 3 dimensions are supported. Received {}.'.format(dim)\n",
-    "            )\n",
-    "\n",
-    "    def forward(self, input):\n",
-    "        \"\"\"\n",
-    "        Apply gaussian filter to input.\n",
-    "        Arguments:\n",
-    "            input (torch.Tensor): Input to apply gaussian filter on.\n",
-    "        Returns:\n",
-    "            filtered (torch.Tensor): Filtered output.\n",
-    "        \"\"\"\n",
-    "        return self.conv(input, weight=self.weight, groups=self.groups)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 18,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "smoothing = GaussianSmoothing(1024, 5, 1, dim=1)\n",
-    "input = torch.rand(4, 16, 100, 64)\n",
-    "b = input.shape[0]\n",
-    "numhead = input.shape[1]\n",
-    "t = input.shape[2]\n",
-    "c = input.shape[3]\n",
-    "input = input.permute(0,1,3,2)\n",
-    "input = input.reshape(b, numhead*c, t)\n",
-    "input = F.pad(input, (2, 2), mode='reflect')\n",
-    "output = smoothing(input)\n",
-    "output = output.reshape(b, numhead, c, t)\n",
-    "output = output.permute(0,1,3,2)"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": 19,
-   "metadata": {},
-   "outputs": [
-    {
-     "data": {
-      "text/plain": [
-       "torch.Size([4, 16, 100, 64])"
-      ]
-     },
-     "execution_count": 19,
-     "metadata": {},
-     "output_type": "execute_result"
-    }
-   ],
-   "source": [
-    "input\n",
-    "output.shape"
-   ]
-  },
-  {
-   "cell_type": "code",
-   "execution_count": null,
-   "metadata": {},
-   "outputs": [],
-   "source": [
-    "input = torch.rand(4, 16, 100, 64)\n",
-    "attention = torch.normal(0,1 size = input.shape)"
-   ]
-  }
- ],
- "metadata": {
-  "kernelspec": {
-   "display_name": "Python 3",
-   "language": "python",
-   "name": "python3"
-  },
-  "language_info": {
-   "codemirror_mode": {
-    "name": "ipython",
-    "version": 3
-   },
-   "file_extension": ".py",
-   "mimetype": "text/x-python",
-   "name": "python",
-   "nbconvert_exporter": "python",
-   "pygments_lexer": "ipython3",
-   "version": "3.8.3"
-  }
- },
- "nbformat": 4,
- "nbformat_minor": 4
-}

scripts/get_incorrect_samples.py

@@ -1,88 +0,0 @@
-r"""Compute active speaker detection performance for the AVA dataset.
-Please send any questions about this code to the Google Group ava-dataset-users:
-https://groups.google.com/forum/#!forum/ava-dataset-users
-Example usage:
-python -O get_ava_active_speaker_performance.py \
--g testdata/eval.csv \
--p testdata/predictions.csv \
--v
-"""
-
-from __future__ import absolute_import
-from __future__ import division
-from __future__ import print_function
-
-import argparse
-import logging
-import time, warnings
-import numpy as np
-import pandas as pd
-import matplotlib.pyplot as plt
-warnings.filterwarnings("ignore")
-
-
-def parse_arguments():
-    """Parses command-line flags.
-  Returns:
-    args: a named tuple containing three file objects args.labelmap,
-    args.groundtruth, and args.detections.
-  """
-    parser = argparse.ArgumentParser()
-    parser.add_argument("-g",
-                        "--groundtruth",
-                        help="CSV file containing ground truth.",
-                        type=argparse.FileType("r"),
-                        required=True)
-    parser.add_argument("-p",
-                        "--predictions",
-                        help="CSV file containing active speaker predictions.",
-                        type=argparse.FileType("r"),
-                        required=True)
-    parser.add_argument("-v", "--verbose", help="Increase output verbosity.", action="store_true")
-    return parser.parse_args()
-
-
-def run_evaluation(groundtruth, predictions):
-    prediction = pd.read_csv(predictions)
-    groundtruth = pd.read_csv(groundtruth)
-    wrong_list = []
-    num = 0
-    audible_num = 0
-    total = 0
-    for i, row in prediction.iterrows():
-        entity_id = row['entity_id']
-        ts = row['frame_timestamp']
-        if row['score'] < 0.5:
-            label = "NOT_SPEAKING"
-        else:
-            label = "SPEAKING_AUDIBLE"
-
-        true_label = groundtruth.loc[(groundtruth['entity_id'] == entity_id) &
-                                     (groundtruth['frame_timestamp'] == ts)].iloc[0]["label"]
-        if true_label != label:
-            wrong_list.append([entity_id, ts, true_label, label])
-
-        if label == "SPEAKING_AUDIBLE":
-            num += 1
-        if true_label == "SPEAKING_AUDIBLE":
-            audible_num += 1
-        total += 1
-    print(num, audible_num, total)
-
-    df = pd.DataFrame(wrong_list, columns=['entity_id', 'frame_timestamp', "gt", "prediction"])
-    df = df.sort_values(by=["frame_timestamp"])
-    df.to_csv("wrong_list.csv")
-
-
-def main():
-    start = time.time()
-    args = parse_arguments()
-    if args.verbose:
-        logging.basicConfig(level=logging.DEBUG)
-    del args.verbose
-    run_evaluation(**vars(args))
-    logging.info("Computed in %s seconds", time.time() - start)
-
-
-if __name__ == "__main__":
-    main()