Merge branch 'master' into ns/benchmarks

.github/workflows/format_pr.yml

@@ -0,0 +1,36 @@
+name: format-pr
+
+on:
+  schedule:
+    - cron: '0 0 * * *'
+jobs:
+  build:
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        julia-version: [1.4.2]
+        julia-arch: [x86]
+        os: [ubuntu-latest]
+    steps:
+      - uses: julia-actions/setup-julia@latest
+        with:
+          version: ${{ matrix.julia-version }}
+
+      - uses: actions/checkout@v2
+      - name: Install JuliaFormatter and format
+        run: |
+          julia  -e 'import Pkg; Pkg.add("JuliaFormatter")'
+          julia  -e 'using JuliaFormatter; format(".")'
+      # https://github.com/marketplace/actions/create-pull-request
+      - name: Create Pull Request
+        uses: peter-evans/create-pull-request@v2
+        with:
+          token: ${{ secrets.GITHUB_TOKEN }}
+          commit-message: Format .jl files
+          title: 'Automatic JuliaFormatter.jl run'
+          branch: auto-juliaformatter-pr
+          labels: formatting, automated pr, no changelog
+      - name: Check outputs
+        run: |
+          echo 'Pull Request Number - ${{ env.PULL_REQUEST_NUMBER }}'
+          echo 'Pull Request Number - ${{ steps.cpr.outputs.pr_number }}'

benchmarks/plot.jl

@@ -11,6 +11,7 @@ function read_bm_transforms(fname, framework)
     )
     for line in eachline(fname)
         raw = split(line)
+<<<<<<< HEAD
         push!(
             data,
             (
@@ -21,6 +22,9 @@ function read_bm_transforms(fname, framework)
                 parse(Float64, raw[5]),
             ),
         )
+=======
+        push!(data, (raw[1], raw[2], parse(Int, raw[3]), parse(Float64, raw[4])))
+>>>>>>> master
     end
     data[!, :framework] .= framework
     return data

benchmarks/transforms.jl

@@ -38,12 +38,15 @@ function realign_point_cloud(npoints)
     return rot(pc)
 end
 
+<<<<<<< HEAD
 function realign_trimesh(_npoints)
     pc = generate_trimesh(_npoints)
     rot = RotateTriMesh(-ROT_MATRIX)
     return rot(pc)
 end
 
+=======
+>>>>>>> master
 ROT_MATRIX = [
     1.0 2.0 3.0
     0.2 0.5 0.9
@@ -75,6 +78,7 @@ gpu_bm_trimesh = setup_benchmark_record(names)
 
 println("DEVICE: CPU")
 for _npoints in npoint_arr
+<<<<<<< HEAD
     pcloud_arr = [
         (ScalePointCloud(0.5; inplace = false), "ScalePointCloud"),
         (RotatePointCloud(ROT_MATRIX; inplace = false), "RotatePointCloud"),
@@ -85,6 +89,25 @@ for _npoints in npoint_arr
                RotatePointCloud(ROT_MATRIX; inplace = false),
                ReAlignPointCloud(realign_point_cloud(_npoints);inplace = false,),
                NormalizePointCloud()),"Chain",)
+=======
+    arr = [
+        (ScalePointCloud(0.5; inplace = false), "ScalePointCloud"),
+        (RotatePointCloud(ROT_MATRIX; inplace = false), "RotatePointCloud"),
+        (
+            ReAlignPointCloud(realign_point_cloud(_npoints); inplace = false),
+            "RealignPointCloud",
+        ),
+        (NormalizePointCloud(inplace = false), "NormalizePointCloud"),
+        (
+            Chain(
+                ScalePointCloud(0.5; inplace = false),
+                RotatePointCloud(ROT_MATRIX; inplace = false),
+                ReAlignPointCloud(realign_point_cloud(_npoints); inplace = false),
+                NormalizePointCloud(),
+            ),
+            "Chain",
+        ),
+>>>>>>> master
     ]
 
     trimesh_arr = [
@@ -123,6 +146,7 @@ if has_cuda()
     println("CUDA is on. Running GPU Benchmarks")
     println("DEVICE: GPU")
     for _npoints in npoint_arr
+<<<<<<< HEAD
         pcloud_arr = [
             (ScalePointCloud(0.5; inplace = false), "ScalePointCloud"),
             (RotatePointCloud(ROT_MATRIX; inplace = false), "RotatePointCloud"),
@@ -162,26 +186,66 @@ if has_cuda()
             generate_trimesh,
             (op, pc) -> (CUDA.@sync op(pc)),
             gpu
+=======
+        arr = [
+            (ScalePointCloud(0.5; inplace = false), "ScalePointCloud"),
+            (RotatePointCloud(ROT_MATRIX; inplace = false), "RotatePointCloud"),
+            (
+                ReAlignPointCloud(realign_point_cloud(_npoints); inplace = false),
+                "RealignPointCloud",
+            ),
+            (NormalizePointCloud(inplace = false), "NormalizePointCloud"),
+            (
+                Chain(
+                    ScalePointCloud(0.5; inplace = false),
+                    RotatePointCloud(ROT_MATRIX; inplace = false),
+                    ReAlignPointCloud(realign_point_cloud(_npoints); inplace = false),
+                    NormalizePointCloud(inplace = false),
+                ),
+                "Chain",
+            ),
+        ]
+
+        println("Running benchmarks for npoints = $_npoints")
+        run_benchmarks!(
+            gpu_benchmarks,
+            arr,
+            _npoints,
+            (op, pc) -> (CuArrays.@sync op(pc)),
+            gpu,
+>>>>>>> master
         )
         println()
     end
 end
 
+<<<<<<< HEAD
 function save_bm(fname, rep, cpu_benchmarks, gpu_benchmarks)
+=======
+function save_bm(fname, cpu_benchmarks, gpu_benchmarks)
+>>>>>>> master
     open(fname, "w") do io
         device = "cpu"
         for (key, values) in cpu_benchmarks
             for (p, v) in zip(npoint_arr, values)
+<<<<<<< HEAD
                 Printf.@printf(io, "%s %s %s %d %f ms\n",
                                rep, device, key, p, v)
+=======
+                Printf.@printf(io, "%s %s %d %f ms\n", device, key, p, v)
+>>>>>>> master
             end
         end
 
         device = "gpu"
         for (key, values) in gpu_benchmarks
             for (p, v) in zip(npoint_arr, values)
+<<<<<<< HEAD
                 Printf.@printf(io, "%s %s %s %d %f ms\n",
                                rep, device, key, p, v)
+=======
+                Printf.@printf(io, "%s %s %d %f ms\n", device, key, p, v)
+>>>>>>> master
             end
         end
     end

benchmarks/triangle_mesh.jl

@@ -3,13 +3,13 @@ using Flux3D, BenchmarkTools
 # mesh = load_trimesh(joinpath(@__DIR__, "../test/meshes/teapot.obj"))
 mesh = load_trimesh("./test/meshes/teapot.obj")
 
-trail = @benchmark $get_edges_packed($mesh; refresh=true)
+trail = @benchmark $get_edges_packed($mesh; refresh = true)
 minimum(trail.times) * 1.0e-6
 
-trail = @benchmark $get_faces_to_edges_packed($mesh; refresh=true)
+trail = @benchmark $get_faces_to_edges_packed($mesh; refresh = true)
 minimum(trail.times) * 1.0e-6
 
-trail = @benchmark $get_laplacian_packed($mesh; refresh=true)
+trail = @benchmark $get_laplacian_packed($mesh; refresh = true)
 minimum(trail.times) * 1.0e-6
 
 trail = @benchmark $compute_verts_normals_packed($mesh)

docs/make.jl

@@ -2,20 +2,16 @@ using Flux3D
 using Documenter
 
 makedocs(;
-    modules=[Flux3D],
-    authors="Nirmal P. Suthar <nirmalps@iitk.ac.in>",
-    repo="https://github.com/nirmal-suthar/Flux3D.jl/blob/{commit}{path}#L{line}",
-    sitename="Flux3D.jl",
-    format=Documenter.HTML(;
-        prettyurls=get(ENV, "CI", "false") == "true",
-        canonical="https://nirmal-suthar.github.io/Flux3D.jl",
-        assets=String[],
+    modules = [Flux3D],
+    authors = "Nirmal P. Suthar <nirmalps@iitk.ac.in>",
+    repo = "https://github.com/nirmal-suthar/Flux3D.jl/blob/{commit}{path}#L{line}",
+    sitename = "Flux3D.jl",
+    format = Documenter.HTML(;
+        prettyurls = get(ENV, "CI", "false") == "true",
+        canonical = "https://nirmal-suthar.github.io/Flux3D.jl",
+        assets = String[],
     ),
-    pages=[
-        "Home" => "index.md",
-    ],
+    pages = ["Home" => "index.md"],
 )
 
-deploydocs(;
-    repo="github.com/nirmal-suthar/Flux3D.jl.git",
-)
+deploydocs(; repo = "github.com/nirmal-suthar/Flux3D.jl.git")

examples/dgcnn_classification.jl

@@ -25,33 +25,56 @@ end
 function get_processed_data(args)
     # Fetching the train and validation data and getting them into proper shape	
     if args.num_classes == 10
-        dset = ModelNet10.dataset(;mode=:pointcloud, npoints=args.npoints, transform=NormalizePointCloud())
+        dset = ModelNet10.dataset(;
+            mode = :pointcloud,
+            npoints = args.npoints,
+            transform = NormalizePointCloud(),
+        )
     elseif args.num_classes == 40
-        dset = ModelNet40.dataset(;mode=:pointcloud, npoints=args.npoints, transform=NormalizePointCloud())
+        dset = ModelNet40.dataset(;
+            mode = :pointcloud,
+            npoints = args.npoints,
+            transform = NormalizePointCloud(),
+        )
     else
         error("ModelNet dataset with $(args.num_classes) is not supported.
                 Currently supported num_classes for ModelNet dataset is {10,40}")
     end
 
-    data = [dset[i].data.points for i in 1:length(dset)]
-    labels = onehotbatch([dset[i].ground_truth for i in 1:length(dset)],1:args.num_classes)
+    data = [dset[i].data.points for i = 1:length(dset)]
+    labels = onehotbatch([dset[i].ground_truth for i = 1:length(dset)], 1:args.num_classes)
 
     #onehot encode labels of batch
-    train = [(cat(data[i]..., dims = 3), labels[:,i]) for i in partition(1:length(data), args.batch_size)] .|> args.device
-
+    train =
+        [
+            (cat(data[i]..., dims = 3), labels[:, i])
+            for i in partition(1:length(data), args.batch_size)
+        ] .|> args.device
+
     if args.num_classes == 10
-        VAL = ModelNet10.dataset(;mode=:pointcloud, train=false, npoints=args.npoints, transform=NormalizePointCloud())
+        VAL = ModelNet10.dataset(;
+            mode = :pointcloud,
+            train = false,
+            npoints = args.npoints,
+            transform = NormalizePointCloud(),
+        )
     elseif args.num_classes == 40
-        VAL = ModelNet40.dataset(;mode=:pointcloud, train=false, npoints=args.npoints, transform=NormalizePointCloud())
+        VAL = ModelNet40.dataset(;
+            mode = :pointcloud,
+            train = false,
+            npoints = args.npoints,
+            transform = NormalizePointCloud(),
+        )
     else
         error("ModelNet dataset with $(args.num_classes) is not supported.
                 Currently supported num_classes for ModelNet dataset is {10,40}")
     end
 
-    valX = cat([VAL[i].data.points for i in 1:length(VAL)]..., dims=3) |> args.device
-    valY = onehotbatch([VAL[i].ground_truth for i in 1:length(VAL)], 1:args.num_classes) |> args.device
+    valX = cat([VAL[i].data.points for i = 1:length(VAL)]..., dims = 3) |> args.device
+    valY =
+        onehotbatch([VAL[i].ground_truth for i = 1:length(VAL)], 1:args.num_classes) |> args.device
 
-    val = (valX,valY)	
+    val = (valX, valY)
     return train, val
 end
 
@@ -67,17 +90,18 @@ function train(; kwargs...)
         args.device = cpu
         @info("Training on CPU")
     end
-    
+
     @info("Loading Data...")
     # Load the train, validation data 
-    train,val = get_processed_data(args)
+    train, val = get_processed_data(args)
 
-    @info("Initializing Model...")	
+    @info("Initializing Model...")
     # Defining the loss and accuracy functions
     m = DGCNN(args.num_classes, args.K, args.npoints) |> args.device
 
     loss(x, y) = crossentropy(m(x), y)
-    accuracy(x, y) = mean(onecold(cpu(m(x)), 1:args.num_classes) .== onecold(cpu(y), 1:args.num_classes))
+    accuracy(x, y) =
+        mean(onecold(cpu(m(x)), 1:args.num_classes) .== onecold(cpu(y), 1:args.num_classes))
 
     ## Training
     opt = ADAM(args.lr)
@@ -90,18 +114,18 @@ end
 
 function custom_train!(loss, ps, data, opt, epochs, (valX, valY), accuracy)
     ps = Zygote.Params(ps)
-    for epoch in 1:epochs
+    for epoch = 1:epochs
         running_loss = 0
         for d in data
-        gs = gradient(ps) do
-            training_loss = loss(d...)
-            running_loss += training_loss
-            return training_loss
-        end
-        Flux.update!(opt, ps, gs)
+            gs = gradient(ps) do
+                training_loss = loss(d...)
+                running_loss += training_loss
+                return training_loss
+            end
+            Flux.update!(opt, ps, gs)
         end
         print("Epoch: $(epoch), epoch_loss: $(running_loss), accuracy: $(accuracy(valX, valY))\n")
     end
 end
 
-m = train()
+m = train()