Fix imread call to use apply_gamma=False instead of ignoregamma=True

Dockerfile

@@ -0,0 +1,4 @@
+FROM ubuntu:20.04
+
+RUN apt-get install build-essential git python3-dev python3-pip libopenexr-dev libxi-dev \
+libglfw3-dev libglew-dev libomp-dev libxinerama-dev libxcursor-dev

MyLog/run1_observation.md

@@ -0,0 +1,19 @@
+# Run1 Observation
+
+## Data
+`image_4/` has higher resolution compared with `images_8/`
+- `images` (Original data) has higher resolution compared with `images_4/`
+- All folders contain images from 00-19. (Though their name is different)
+Train data: [ 1  2  3  4  5  6  7  9 10 11 12 13 14 15 17 18 19]
+Test and Val data: [0 8 16]
+
+## Training - `train()` (from line 575)
+- `nerf` model is created and define in `create_nerf(args)` function.
+
+- main part are on the `Core optimization loop`.
+
+## Nerf: `create_nerf(args)` (from line 378): initialize the NeRE's MLP model
+- `init_nerf_model` (from `run_nerf_helpers.py`)
+
+
+

dockerfiles_old/Dockerfile

@@ -0,0 +1,54 @@
+# Successfully built dockerfile to use Nsight Systems on Ubuntu 20.04 with pyenv and pyenv-virtualenv
+FROM nvidia/cuda:12.3.2-cudnn9-devel-ubuntu20.04
+
+RUN apt-get update && \
+    apt-get install -y wget
+# RUN git --version
+
+# ## Install Nsight Systems
+RUN wget https://developer.nvidia.com/downloads/assets/tools/secure/nsight-systems/2024_4/NsightSystems-linux-cli-public-2024.4.1.61-3431596.deb
+
+# # Set the timezone
+RUN ln -sf /usr/share/zoneinfo/Asia/Tokyo /etc/localtime 
+
+RUN apt-get install -y ./NsightSystems-linux-cli-public-2024.4.1.61-3431596.deb
+
+RUN rm ./NsightSystems-linux-cli-public-2024.4.1.61-3431596.deb && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN sysctl -w kernel.perf_event_paranoid=1
+
+# ## Install basic tools (list of necessary tools are taken from https://qiita.com/nyakiri_0726/items/a33f404b5e1be9352b85)
+# `apt-get update` is necessary before `apt-get install` to avoid error
+# RUN apt-get update && apt-get install -y make build-essential git vim zlib1g-dev
+RUN apt-get update && apt-get install -y build-essential git vim libssl-dev zlib1g-dev libbz2-dev libreadline-dev \
+    libsqlite3-dev wget curl llvm libncurses5-dev libncursesw5-dev xz-utils tk-dev libffi-dev \
+    liblzma-dev python-openssl git vim less
+
+# # Environment variables
+ENV HOME /root
+ENV PYENV_ROOT $HOME/.pyenv
+ENV PATH $PYENV_ROOT/bin:$PATH
+ARG PYTHON_VERSION=3.7.17
+
+RUN git clone https://github.com/pyenv/pyenv.git ~/.pyenv
+
+RUN echo 'export PATH="$HOME/.pyenv/bin:$PATH"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv init --path)"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv init -)"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv virtualenv-init -)"' >> ~/.bashrc && \
+    echo 'export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python' >> ~/.bashrc
+
+# it may takes time in `patching file setup.py`
+RUN pyenv install 3.7.17
+
+# Install pyenv-virtualenv
+# NOTE: it should not be `$(PYENV_ROOT)/plugins/pyenv-virtualenv` but `$PYENV_ROOT/plugins/pyenv-virtualenv`
+RUN git clone https://github.com/pyenv/pyenv-virtualenv.git $PYENV_ROOT/plugins/pyenv-virtualenv
+RUN echo 'eval "$(pyenv virtualenv-init -)"' >> ~/.bashrc
+# RUN source ~/.bashrc
+RUN pyenv virtualenv 3.7.17 nerf_env
+
+
+CMD [ "/bin/bash" ]

dockerfiles_old/Dockerfile copy

@@ -0,0 +1,49 @@
+FROM nvidia/cuda:12.3.2-cudnn9-devel-ubuntu20.04
+
+RUN apt-get update && \
+    apt-get install -y wget
+
+## Install Nsight Systems
+RUN wget https://developer.nvidia.com/downloads/assets/tools/secure/nsight-systems/2024_4/NsightSystems-linux-cli-public-2024.4.1.61-3431596.deb
+
+# Set the timezone
+RUN ln -sf /usr/share/zoneinfo/Asia/Tokyo /etc/localtime 
+
+RUN apt-get install -y ./NsightSystems-linux-cli-public-2024.4.1.61-3431596.deb
+
+RUN rm ./NsightSystems-linux-cli-public-2024.4.1.61-3431596.deb && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN sysctl -w kernel.perf_event_paranoid=1
+
+## Install basic tools
+RUN apt-get install -y make build-essential
+# RUN curl https://pyenv.run | bash
+# RUN pyenv install 3.7.17
+
+RUN echo 'export PATH="$HOME/.pyenv/bin:$PATH"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv init --path)"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv init -)"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv virtualenv-init -)"' >> ~/.bashrc && \
+    echo 'export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python' >> ~/.bashrc
+
+# RUN /root/.pyenv/bin/pyenv install 3.7.17 && \
+#     /root/.pyenv/bin/pyenv global 3.7.17
+
+RUN wget https://www.python.org/ftp/python/3.7.17/Python-3.7.17.tgz && \
+    tar -xzf Python-3.7.17.tgz
+
+RUN cd Python-3.7.17 && \
+    ./configure --enable-optimizations && \
+    make -j$(nproc) && \
+    make altinstall
+
+RUN rm -rf Python-3.7.17 Python-3.7.17.tgz && \
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+RUN update-alternatives --install /usr/bin/python3 python3 /usr/local/bin/python3.7 1 && \
+    update-alternatives --install /usr/bin/pip3 pip3 /usr/local/bin/pip3.7 1
+
+CMD [ "/bin/bash" ]

dockerfiles_old/Dockerfile_torchprofiler

@@ -0,0 +1,41 @@
+FROM ubuntu:20.04
+
+ENV HOME /root
+ENV PYENV_ROOT $HOME/.pyenv
+ENV PATH $PYENV_ROOT/bin:$PATH
+
+# # Set the timezone
+RUN ln -sf /usr/share/zoneinfo/Asia/Tokyo /etc/localtime 
+
+RUN apt-get update && apt-get install -y wget build-essential libssl-dev zlib1g-dev libbz2-dev libreadline-dev \
+    libsqlite3-dev wget curl llvm libncurses5-dev libncursesw5-dev xz-utils tk-dev libffi-dev \
+    liblzma-dev python-openssl git vim less
+
+
+RUN git clone https://github.com/pyenv/pyenv.git ~/.pyenv
+
+RUN echo 'export PATH="$HOME/.pyenv/bin:$PATH"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv init --path)"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv init -)"' >> ~/.bashrc && \
+    echo 'eval "$(pyenv virtualenv-init -)"' >> ~/.bashrc && \
+    echo 'export PROTOCOL_BUFFERS_PYTHON_IMPLEMENTATION=python' >> ~/.bashrc
+
+# it may takes time in `patching file setup.py`
+RUN pyenv install 3.7.17
+
+# Install pyenv-virtualenv
+# NOTE: it should not be `$(PYENV_ROOT)/plugins/pyenv-virtualenv` but `$PYENV_ROOT/plugins/pyenv-virtualenv`
+RUN git clone https://github.com/pyenv/pyenv-virtualenv.git $PYENV_ROOT/plugins/pyenv-virtualenv
+RUN echo 'eval "$(pyenv virtualenv-init -)"' >> ~/.bashrc
+# RUN source ~/.bashrc
+RUN pyenv virtualenv 3.7.17 nerf_env
+
+# RUN cd $PYENV_ROOT && \
+# 		src/configure && \
+# 		make -C src
+
+RUN . $HOME/.bashrc
+RUN pyenv activate nerf_env
+RUN pip install -r requirements.txt
+RUN pip install torchprofiler
+

load_llff.py

@@ -107,7 +107,7 @@ def _load_data(basedir, factor=None, width=None, height=None, load_imgs=True):
 
     def imread(f):
         if f.endswith('png'):
-            return imageio.imread(f, ignoregamma=True)
+            return imageio.imread(f, apply_gamma=False)
         else:
             return imageio.imread(f)
 

requirements.txt

@@ -0,0 +1,7 @@
+tensorflow-gpu==1.15
+numpy
+matplotlib
+imageio
+imageio-ffmpeg
+configargparse
+imagemagick

run_nerf.py

@@ -742,8 +742,7 @@ def train():
         np.random.shuffle(rays_rgb)
         print('done')
         i_batch = 0
-
-    N_iters = 1000000
+    N_iters = 3
     print('Begin')
     print('TRAIN views are', i_train)
     print('TEST views are', i_test)
@@ -802,16 +801,24 @@ def train():
                 batch_rays = tf.stack([rays_o, rays_d], 0)
                 target_s = tf.gather_nd(target, select_inds)
 
+        time1 = time.time()
+        ray_time = time1 - time0
+
         #####  Core optimization loop  #####
 
         with tf.GradientTape() as tape:
 
             # Make predictions for color, disparity, accumulated opacity.
+            # NOTE: これがメインの処理: 3 -> 4に相当
+            time_before_render = time.time()
             rgb, disp, acc, extras = render(
                 H, W, focal, chunk=args.chunk, rays=batch_rays,
                 verbose=i < 10, retraw=True, **render_kwargs_train)
+            time_after_render = time.time()
+            time_render = time_after_render - time_before_render
 
             # Compute MSE loss between predicted and true RGB.
+            time_before_loss = time.time()
             img_loss = img2mse(rgb, target_s)
             trans = extras['raw'][..., -1]
             loss = img_loss
@@ -822,9 +829,15 @@ def train():
                 img_loss0 = img2mse(extras['rgb0'], target_s)
                 loss += img_loss0
                 psnr0 = mse2psnr(img_loss0)
+
+            time_after_loss = time.time()
+            time_calc_loss = time_after_loss - time_before_loss
 
+        time2 = time.time()
         gradients = tape.gradient(loss, grad_vars)
         optimizer.apply_gradients(zip(gradients, grad_vars))
+        time3 = time.time()
+        time_backprop = time3 - time2
 
         dt = time.time()-time0
 
@@ -874,7 +887,9 @@ def save_weights(net, prefix, i):
         if i % args.i_print == 0 or i < 10:
 
             print(expname, i, psnr.numpy(), loss.numpy(), global_step.numpy())
-            print('iter time {:.05f}'.format(dt))
+            print('iter time {:.05f} (ray calc: {:.05f} ({:.02f}%), rendering: {:.05f} ({:.02f}%), loss: {:.05f} ({:.02f}%), backprop: {:.05f} ({:.02f}%))'.format(dt, ray_time, ray_time/dt*100, time_render, time_render/dt*100, time_calc_loss, time_calc_loss/dt*100, time_backprop, time_backprop/dt*100))
+
+
             with tf.contrib.summary.record_summaries_every_n_global_steps(args.i_print):
                 tf.contrib.summary.scalar('loss', loss)
                 tf.contrib.summary.scalar('psnr', psnr)

sample.cu

@@ -0,0 +1,18 @@
+#include "stdio.h"
+__global__ void add(int a, int b, int *c)
+{
+    *c = a + b;
+}
+int main()
+{
+    int a, b, c;
+    int *dev_c;
+    a = 3;
+    b = 4;
+    cudaMalloc((void **)&dev_c, sizeof(int));
+    add<<<1, 1>>>(a, b, dev_c);
+    cudaMemcpy(&c, dev_c, sizeof(int), cudaMemcpyDeviceToHost);
+    printf("%d + %d is %d\n", a, b, c);
+    cudaFree(dev_c);
+    return 0;
+}

test.py

@@ -0,0 +1,29 @@
+import torch
+import torch.nn as nn
+
+# Define a simple neural network
+class SimpleNet(nn.Module):
+    def __init__(self):
+        super(SimpleNet, self).__init__()
+        self.conv1 = nn.Conv2d(in_channels=3, out_channels=16, kernel_size=3, stride=1, padding=1)
+        self.conv2 = nn.Conv2d(in_channels=16, out_channels=32, kernel_size=3, stride=1, padding=1)
+        self.fc1 = nn.Linear(32 * 8 * 8, 128)
+        self.fc2 = nn.Linear(128, 10)
+
+    def forward(self, x):
+        x = torch.relu(self.conv1(x))
+        x = torch.relu(self.conv2(x))
+        x = torch.flatten(x, 1)
+        x = torch.relu(self.fc1(x))
+        x = self.fc2(x)
+        return x
+
+# Create a model instance
+model = SimpleNet()
+
+# Create a dummy input tensor
+dummy_input = torch.randn(1, 3, 8, 8)
+
+# Profile the FLOPS
+output = model(dummy_input)
+print("Finished!")

test2.py

@@ -0,0 +1,34 @@
+import torch
+import torch.nn as nn
+import torchprofile
+
+# Define a simple neural network
+class SimpleNet(nn.Module):
+    def __init__(self):
+        super(SimpleNet, self).__init__()
+        self.conv1 = nn.Conv2d(in_channels=3, out_channels=16, kernel_size=3, stride=1, padding=1)
+        self.conv2 = nn.Conv2d(in_channels=16, out_channels=32, kernel_size=3, stride=1, padding=1)
+        self.fc1 = nn.Linear(32 * 8 * 8, 128)
+        self.fc2 = nn.Linear(128, 10)
+
+    def forward(self, x):
+        x = torch.relu(self.conv1(x))
+        x = torch.relu(self.conv2(x))
+        x = torch.flatten(x, 1)
+        x = torch.relu(self.fc1(x))
+        x = self.fc2(x)
+        return x
+
+# Create a model instance
+model = SimpleNet()
+
+# Create a dummy input tensor
+dummy_input = torch.randn(1, 3, 8, 8)
+
+# Profile the FLOPS
+with torchprofile.profile(model, dummy_input) as profiler:
+    output = model(dummy_input)
+
+# Print the profiling results
+print(profiler)
+

test3.py

@@ -0,0 +1,33 @@
+import torch
+import torch.nn as nn
+from torchprofile import profile  # Import the profile function
+
+# Define a simple neural network
+class SimpleNet(nn.Module):
+    def __init__(self):
+        super(SimpleNet, self).__init__()
+        self.conv1 = nn.Conv2d(in_channels=3, out_channels=16, kernel_size=3, stride=1, padding=1)
+        self.conv2 = nn.Conv2d(in_channels=16, out_channels=32, kernel_size=3, stride=1, padding=1)
+        self.fc1 = nn.Linear(32 * 8 * 8, 128)
+        self.fc2 = nn.Linear(128, 10)
+
+    def forward(self, x):
+        x = torch.relu(self.conv1(x))
+        x = torch.relu(self.conv2(x))
+        x = torch.flatten(x, 1)
+        x = torch.relu(self.fc1(x))
+        x = self.fc2(x)
+        return x
+
+# Create a model instance
+model = SimpleNet()
+
+# Create a dummy input tensor
+dummy_input = torch.randn(1, 3, 8, 8)
+
+# Profile the FLOPS
+with profile(model, dummy_input) as profiler:
+  output = model(dummy_input)
+
+# Print the profiling results
+print(profiler)

test4.py

@@ -0,0 +1,13 @@
+import tensorflow as tf
+
+# Initialize profiler
+tf.profiler.experimental.start(logdir='path/to/logs')
+
+# Define and run your TensorFlow operations
+a = tf.constant(1)
+b = tf.constant(2)
+c = a + b
+print(c)
+
+# Stop profiler
+tf.profiler.experimental.stop()