diff --git a/.gitignore b/.gitignore
index d3df6a7..7329f24 100644
--- a/.gitignore
+++ b/.gitignore
@@ -2,6 +2,7 @@
 # will have compiled files and executables
 debug/
 target/
+venv
 
 # Remove Cargo.lock from gitignore if creating an executable, leave it for libraries
 # More information here https://doc.rust-lang.org/cargo/guide/cargo-toml-vs-cargo-lock.html
diff --git a/Cargo.toml b/Cargo.toml
index 587e6f7..f5b358a 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "camera-intrinsic-calibration"
-version = "0.5.6"
+version = "0.6.0"
 edition = "2021"
 authors = ["Powei Lin <poweilin1994@gmail.com>"]
 readme = "README.md"
@@ -21,13 +21,13 @@ exclude = [
 ]
 
 [dependencies]
-aprilgrid = "0.5.3"
+aprilgrid = "0.6.0"
 camera-intrinsic-model = "0.3.1"
 clap = { version = "4.5.31", features = ["derive"] }
 colorous = "1.0.15"
 env_logger = "0.11.6"
-faer = "0.20.2"
-glam = "0.29.2"
+faer = "0.21.5"
+glam = "0.30.0"
 glob = "0.3.2"
 image = "0.25.5"
 indicatif = { version = "0.17.11", features = ["rayon"] }
@@ -39,9 +39,9 @@ rayon = "1.10.0"
 rerun = "0.17.0"
 serde = { version = "1.0.218", features = ["derive"] }
 serde_json = "1.0.139"
-sqpnp_simple = "0.1.5"
+sqpnp_simple = "0.1.6"
 time = "0.3.37"
-tiny-solver = "0.14.2"
+tiny-solver = "0.17.0"
 
 [[bin]]
 name = "ccrs"
diff --git a/README.md b/README.md
index d0769c8..eee6329 100644
--- a/README.md
+++ b/README.md
@@ -1,12 +1,16 @@
 # camera-intrinsic-calibration
 [![crate](https://img.shields.io/crates/v/camera-intrinsic-calibration.svg)](https://crates.io/crates/camera-intrinsic-calibration)
+[![PyPI - Version](https://img.shields.io/pypi/v/camera-intrinsic-calibration.svg)](https://pypi.org/project/camera-intrinsic-calibration)
+[![PyPI - Python Version](https://img.shields.io/pypi/pyversions/camera-intrinsic-calibration.svg)](https://pypi.org/project/camera-intrinsic-calibration)
 
 A pure rust camera intrinsic calibration library.
 
 ## Installation
 ```sh
-# install cli
+# cargo install cli
 cargo install camera-intrinsic-calibration
+# pip install cli
+pip install camera-intrinsic-calibration
 ```
 Or download from the latest [release](https://github.com/powei-lin/camera-intrinsic-calibration-rs/releases).
 
@@ -26,7 +30,11 @@ ccrs dataset-calib-cam1_1024_16 --model eucm
 ```
 ### Visualize details after calibration
 ```sh
+# use cargo to install rerun
 cargo install rerun-cli --version 0.17.0
+# or use pip to install rerun
+pip install rerun-sdk==0.17.0
+# visualize result
 rerun results/20YYMMDD_HH_MM_SS/logging.rrd
 ```
 <img src="data/rerun_logs.jpg" width="800" alt="example detection">
diff --git a/pyproject.toml b/pyproject.toml
new file mode 100644
index 0000000..ab1d285
--- /dev/null
+++ b/pyproject.toml
@@ -0,0 +1,26 @@
+[build-system]
+requires = ["maturin>=1.4,<2.0"]
+build-backend = "maturin"
+
+[project]
+name = "camera-intrinsic-calibration"
+requires-python = ">=3.8"
+classifiers = [
+    "Programming Language :: Rust",
+    "Programming Language :: Python :: Implementation :: CPython",
+    "Programming Language :: Python :: Implementation :: PyPy",
+    "Programming Language :: Python :: 3.8",
+    "Programming Language :: Python :: 3.9",
+    "Programming Language :: Python :: 3.10",
+    "Programming Language :: Python :: 3.11",
+    "Programming Language :: Python :: 3.12",
+]
+authors = [
+  { name = "Powei Lin", email = "poweilin1994@gmail.com" },
+]
+dynamic = ["version"]
+
+[tool.maturin]
+bindings = "bin"
+module-name = "ccrs"
+strip = true
\ No newline at end of file
diff --git a/scripts/build_macos.sh b/scripts/build_macos.sh
index 4137bb2..b391a2e 100755
--- a/scripts/build_macos.sh
+++ b/scripts/build_macos.sh
@@ -2,4 +2,5 @@ cargo build -r
 cp ./target/release/ccrs ./ccrs
 tar czvf ccrs-aarch64-apple-darwin.tar.gz ./ccrs
 shasum -a 256 ccrs-aarch64-apple-darwin.tar.gz > ccrs-aarch64-apple-darwin.tar.gz.sha256
-rm ./ccrs
\ No newline at end of file
+rm ./ccrs
+maturin build -r
\ No newline at end of file
diff --git a/scripts/cross_build_linux_aarch64.sh b/scripts/cross_build_linux_aarch64.sh
index 9f68351..e1b4464 100755
--- a/scripts/cross_build_linux_aarch64.sh
+++ b/scripts/cross_build_linux_aarch64.sh
@@ -2,4 +2,5 @@ cargo zigbuild --release --target=aarch64-unknown-linux-gnu
 cp ./target/aarch64-unknown-linux-gnu/release/ccrs ./ccrs
 tar czvf ccrs-aarch64-unknown-linux-gnu.tar.gz ./ccrs
 shasum -a 256 ccrs-aarch64-unknown-linux-gnu.tar.gz > ccrs-aarch64-unknown-linux-gnu.tar.gz.sha256
-rm ./ccrs
\ No newline at end of file
+rm ./ccrs
+maturin build -r --target aarch64-unknown-linux-gnu --zig
\ No newline at end of file
diff --git a/scripts/cross_build_linux_x86_64.sh b/scripts/cross_build_linux_x86_64.sh
index e72a2a6..85160fc 100755
--- a/scripts/cross_build_linux_x86_64.sh
+++ b/scripts/cross_build_linux_x86_64.sh
@@ -2,4 +2,5 @@ cargo zigbuild --release --target=x86_64-unknown-linux-gnu
 cp ./target/x86_64-unknown-linux-gnu/release/ccrs ./ccrs
 tar czvf ccrs-x86_64-unknown-linux-gnu.tar.gz ./ccrs
 shasum -a 256 ccrs-x86_64-unknown-linux-gnu.tar.gz > ccrs-x86_64-unknown-linux-gnu.tar.gz.sha256
-rm ./ccrs
\ No newline at end of file
+rm ./ccrs
+maturin build -r --target x86_64-unknown-linux-gnu --zig
\ No newline at end of file
diff --git a/scripts/cross_build_windows_x86_64.sh b/scripts/cross_build_windows_x86_64.sh
index ca21b22..4161b1e 100755
--- a/scripts/cross_build_windows_x86_64.sh
+++ b/scripts/cross_build_windows_x86_64.sh
@@ -2,4 +2,5 @@ cargo zigbuild --release --target=x86_64-pc-windows-gnu
 cp ./target/x86_64-pc-windows-gnu/release/ccrs.exe ./ccrs.exe
 zip ccrs-x86_64-pc-windows-gnu.zip ./ccrs.exe
 shasum -a 256 ccrs-x86_64-pc-windows-gnu.zip > ccrs-x86_64-pc-windows-gnu.zip.sha256
-rm ./ccrs.exe
\ No newline at end of file
+rm ./ccrs.exe
+maturin build -r --target x86_64-pc-windows-gnu --zig
\ No newline at end of file
diff --git a/scripts/setup_cross_compile.sh b/scripts/setup_cross_compile.sh
index 3071f65..913d6a0 100755
--- a/scripts/setup_cross_compile.sh
+++ b/scripts/setup_cross_compile.sh
@@ -1,4 +1,5 @@
 rustup target add aarch64-unknown-linux-gnu
 rustup target add x86_64-unknown-linux-gnu
+rustup target add x86_64-pc-windows-gnu
 cargo install cargo-zigbuild
 brew install zig
\ No newline at end of file
diff --git a/src/data_loader.rs b/src/data_loader.rs
index c9f2bb7..b17f321 100644
--- a/src/data_loader.rs
+++ b/src/data_loader.rs
@@ -88,7 +88,7 @@ pub fn load_euroc(
             let mut sorted_path: Vec<std::path::PathBuf> =
                 img_paths.into_iter().filter_map(img_filter).collect();
 
-            sorted_path.sort_by(|a, b| a.cmp(b));
+            sorted_path.sort();
             let new_paths: Vec<_> = sorted_path.iter().skip(start_idx).step_by(step).collect();
             let mut time_frame: Vec<_> = new_paths
                 .iter()
@@ -137,7 +137,7 @@ pub fn load_others(
             let mut sorted_path: Vec<std::path::PathBuf> =
                 img_paths.into_iter().filter_map(img_filter).collect();
 
-            sorted_path.sort_by(|a, b| a.cmp(b));
+            sorted_path.sort();
             let new_paths: Vec<_> = sorted_path
                 .iter()
                 .skip(start_idx)
diff --git a/src/optimization/homography.rs b/src/optimization/homography.rs
index 0627a2c..6117d12 100644
--- a/src/optimization/homography.rs
+++ b/src/optimization/homography.rs
@@ -1,4 +1,4 @@
-use faer::prelude::SpSolverLstsq;
+use faer::linalg::solvers::SolveLstsqCore;
 use log::debug;
 use nalgebra as na;
 use rand::seq::SliceRandom;
@@ -13,18 +13,18 @@ fn h6_l1l2_solver(six_pt_pairs: &[(glam::Vec2, glam::Vec2)]) -> Option<(f32, na:
         let x_p = pt1.x;
         let y_p = pt1.y;
         unsafe {
-            m1.write_unchecked(r, 0, -x * y_p);
-            m1.write_unchecked(r, 1, -y * y_p);
-            m1.write_unchecked(r, 2, -y_p);
-            m1.write_unchecked(r, 3, x * x_p);
-            m1.write_unchecked(r, 4, x_p * y);
-            m1.write_unchecked(r, 5, x_p);
-            m1.write_unchecked(r, 6, -x * x * y_p - y * y * y_p);
-            m1.write_unchecked(r, 7, x * x * x_p + x_p * y * y);
+            *m1.get_mut_unchecked(r, 0) = -x * y_p;
+            *m1.get_mut_unchecked(r, 1) = -y * y_p;
+            *m1.get_mut_unchecked(r, 2) = -y_p;
+            *m1.get_mut_unchecked(r, 3) = x * x_p;
+            *m1.get_mut_unchecked(r, 4) = x_p * y;
+            *m1.get_mut_unchecked(r, 5) = x_p;
+            *m1.get_mut_unchecked(r, 6) = -x * x * y_p - y * y * y_p;
+            *m1.get_mut_unchecked(r, 7) = x * x * x_p + x_p * y * y;
         }
     }
     // let q_mat = mm1.transpose().qr().q();
-    let q_mat = m1.transpose().qr().compute_q();
+    let q_mat = m1.transpose().qr().compute_Q();
     let q_mat_t = q_mat.transpose();
     let n = q_mat_t.subrows(6, 2);
     let n02 = *n.get(0, 2);
@@ -62,7 +62,7 @@ fn h6_l1l2_solver(six_pt_pairs: &[(glam::Vec2, glam::Vec2)]) -> Option<(f32, na:
     for which_gamma in 0..2 {
         let gamma = g_result[which_gamma];
         let l = -(gamma * n06 + n16) / (-gamma * n02 - n12);
-        let v1 = gamma * n.row(0) + n.row(1);
+        let v1 = faer::Scale(gamma) * n.row(0) + n.row(1);
         temp_h[which_gamma][(0, 0)] = *v1.get(0);
         temp_h[which_gamma][(0, 1)] = *v1.get(1);
         temp_h[which_gamma][(0, 2)] = *v1.get(2);
@@ -79,12 +79,10 @@ fn h6_l1l2_solver(six_pt_pairs: &[(glam::Vec2, glam::Vec2)]) -> Option<(f32, na:
             let x_p = pt1.x;
             let y_p = pt1.y;
             unsafe {
-                eq10a.write_unchecked(row, 0, -x * x_p);
-                eq10a.write_unchecked(row, 1, -x_p * y);
-                eq10a.write_unchecked(row, 2, -l * x * x * x_p - l * x_p * y * y - x_p);
-                eq10a.write_unchecked(
-                    row,
-                    3,
+                *eq10a.get_mut_unchecked(row, 0) = -x * x_p;
+                *eq10a.get_mut_unchecked(row, 1) = -x_p * y;
+                *eq10a.get_mut_unchecked(row, 2) = -l * x * x * x_p - l * x_p * y * y - x_p;
+                *eq10a.get_mut_unchecked(row, 3) =
                     l * x * x * x_p * x_p * temp_h[which_gamma][(0, 2)]
                         + l * x * x * y_p * y_p * temp_h[which_gamma][(0, 2)]
                         + l * x_p * x_p * y * y * temp_h[which_gamma][(0, 2)]
@@ -94,25 +92,21 @@ fn h6_l1l2_solver(six_pt_pairs: &[(glam::Vec2, glam::Vec2)]) -> Option<(f32, na:
                         + x_p * x_p * y * temp_h[which_gamma][(0, 1)]
                         + x_p * x_p * temp_h[which_gamma][(0, 2)]
                         + y * y_p * y_p * temp_h[which_gamma][(0, 1)]
-                        + y_p * y_p * temp_h[which_gamma][(0, 2)],
-                );
+                        + y_p * y_p * temp_h[which_gamma][(0, 2)];
 
-                eq10b.write_unchecked(
-                    row,
-                    0,
-                    -l * x * x * temp_h[which_gamma][(0, 2)]
-                        - l * y * y * temp_h[which_gamma][(0, 2)]
-                        - x * temp_h[which_gamma][(0, 0)]
-                        - y * temp_h[which_gamma][(0, 1)]
-                        - temp_h[which_gamma][(0, 2)],
-                );
+                *eq10b.get_mut_unchecked(row, 0) = -l * x * x * temp_h[which_gamma][(0, 2)]
+                    - l * y * y * temp_h[which_gamma][(0, 2)]
+                    - x * temp_h[which_gamma][(0, 0)]
+                    - y * temp_h[which_gamma][(0, 1)]
+                    - temp_h[which_gamma][(0, 2)];
             }
         }
         // std::cout << "svd\n";
-        let eq10x = eq10a.qr().solve_lstsq(eq10b);
-        // Eigen::JacobiSVD<Eigen::Matrix<float, 6, 4>> svd(
-        //     eq10A, Eigen::ComputeFullU | Eigen::ComputeFullV);
-        // Eigen::Matrix<float, 4, 1> eq10x = svd.solve(eq10b);
+        let mut eq10x = eq10b;
+        eq10a
+            .qr()
+            .solve_lstsq_in_place_with_conj(faer::Conj::No, eq10x.as_mut());
+
         temp_h[which_gamma][(2, 0)] = *eq10x.get(0, 0);
         temp_h[which_gamma][(2, 1)] = *eq10x.get(1, 0);
         temp_h[which_gamma][(2, 2)] = *eq10x.get(2, 0);
diff --git a/src/util.rs b/src/util.rs
index ecd81e0..64aaa93 100644
--- a/src/util.rs
+++ b/src/util.rs
@@ -241,7 +241,7 @@ pub fn convert_model(
         HashMap::<String, na::DVector<f64>>::from([("params".to_string(), target_params_init)]);
 
     // initialize optimizer
-    let optimizer = tiny_solver::LevenbergMarquardtOptimizer::default();
+    let optimizer = tiny_solver::GaussNewtonOptimizer::default();
 
     // distortion parameter bound
     set_problem_parameter_bound("params", &mut problem, target_model, false);
@@ -311,7 +311,7 @@ pub fn init_ucm(
     ]);
 
     // initialize optimizer
-    let optimizer = tiny_solver::LevenbergMarquardtOptimizer::default();
+    let optimizer = tiny_solver::GaussNewtonOptimizer::default();
     if fixed_focal {
         init_focal_alpha_problem.fix_variable("params", 0);
     }
@@ -346,7 +346,7 @@ pub fn init_ucm(
                 0,
                 fixed_focal,
             )
-            .unwrap()
+            .expect("The initial UCM model fitting failed. Might be wrong board configuration.")
             .0,
         )
     } else {
@@ -406,7 +406,7 @@ pub fn calib_camera(
         }
     }
 
-    let optimizer = tiny_solver::LevenbergMarquardtOptimizer::default();
+    let optimizer = tiny_solver::GaussNewtonOptimizer::default();
     // let initial_values = optimizer.optimize(&problem, &initial_values, None);
 
     set_problem_parameter_bound("params", &mut problem, generic_camera, xy_same_focal);
@@ -512,7 +512,7 @@ pub fn init_camera_extrinsic(cam_rtvecs: &[HashMap<usize, RvecTvec>]) -> Vec<Rve
                 ("tvec".to_string(), tvec),
             ]);
 
-            let optimizer = tiny_solver::LevenbergMarquardtOptimizer::default();
+            let optimizer = tiny_solver::GaussNewtonOptimizer::default();
             let result = optimizer.optimize(&problem, &initial_values, None).unwrap();
             println!("extrinsic cam{} cam0", cam_i);
             println!("rvec: {}", result["rvec"]);
@@ -623,7 +623,7 @@ pub fn calib_all_camera_with_extrinsics(
         println!("set focal");
         problem.fix_variable("params0", 0);
     }
-    let optimizer = tiny_solver::LevenbergMarquardtOptimizer::default();
+    let optimizer = tiny_solver::GaussNewtonOptimizer::default();
 
     let result_option = optimizer.optimize(&problem, &initial_values, None);
     if let Some(mut result) = result_option {