Fix visualization of GridEncoding outputs

NVlabs · Feb 11, 2022 · 09f3597 · 09f3597
1 parent 4929074
commit 09f3597
Show file tree

Hide file tree

Showing 3 changed files with 8 additions and 8 deletions.
diff --git a/dependencies/tiny-cuda-nn b/dependencies/tiny-cuda-nn
diff --git a/include/neural-graphics-primitives/nerf_network.h b/include/neural-graphics-primitives/nerf_network.h
@@ -180,7 +180,7 @@ class NerfNetwork : public tcnn::Network<float, T> {
 		}
 		m_density_network.reset(tcnn::create_network<T>(local_density_network_config));
 
-		m_rgb_network_input_width = tcnn::next_multiple(m_dir_encoding->num_encoded_dims() + m_density_network->padded_output_width() - 1, rgb_alignment);
+		m_rgb_network_input_width = tcnn::next_multiple(m_dir_encoding->num_encoded_dims() + m_density_network->padded_output_width(), rgb_alignment);
 
 		json local_rgb_network_config = rgb_network;
 		local_rgb_network_config["n_input_dims"] = m_rgb_network_input_width;
@@ -509,11 +509,11 @@ class NerfNetwork : public tcnn::Network<float, T> {
 
 	uint32_t width(uint32_t layer) const override {
 		if (layer == 0) {
-			return m_forward.density_network_input.m();
+			return m_pos_encoding->num_encoded_dims();
 		} else if (layer < m_density_network->num_forward_activations() + 1) {
 			return m_density_network->width(layer - 1);
 		} else if (layer == m_density_network->num_forward_activations() + 1) {
-			return m_forward.rgb_network_input.m();
+			return m_rgb_network_input_width;
 		} else {
 			return m_rgb_network->width(layer - 2 - m_density_network->num_forward_activations());
 		}
@@ -523,17 +523,17 @@ class NerfNetwork : public tcnn::Network<float, T> {
 		return m_density_network->num_forward_activations() + m_rgb_network->num_forward_activations() + 2;
 	}
 
-	const T* forward_activations(uint32_t layer) const override {
+	std::pair<const T*, tcnn::MatrixLayout> forward_activations(uint32_t layer) const override {
 		if (!m_forward.density_network_input.data()) {
 			throw std::runtime_error{"Must call forward() before accessing activations."};
 		}
 
 		if (layer == 0) {
-			return m_forward.density_network_input.data();
+			return {m_forward.density_network_input.data(), m_pos_encoding->output_layout()};
 		} else if (layer < m_density_network->num_forward_activations() + 1) {
 			return m_density_network->forward_activations(layer - 1);
 		} else if (layer == m_density_network->num_forward_activations() + 1) {
-			return m_forward.rgb_network_input.data();
+			return {m_forward.rgb_network_input.data(), m_dir_encoding->output_layout()};
 		} else {
 			return m_rgb_network->forward_activations(layer - 2 - m_density_network->num_forward_activations());
 		}

diff --git a/src/testbed_sdf.cu b/src/testbed_sdf.cu
@@ -843,7 +843,7 @@ void Testbed::render_sdf(
 	if (m_render_mode == ERenderMode::Slice) {
 		if (m_visualized_dimension == -1) {
 			distance_function(n_hit, rays_hit.pos, rays_hit.distance, stream);
-			extract_dimension_pos_neg_kernel<float><<<n_blocks_linear(n_hit*3), n_threads_linear, 0, stream>>>(n_hit*3, 0, 1, 3, rays_hit.distance.data(), (float*)rays_hit.normal.data());
+			extract_dimension_pos_neg_kernel<float><<<n_blocks_linear(n_hit*3), n_threads_linear, 0, stream>>>(n_hit*3, 0, 1, 3, rays_hit.distance.data(), CM, (float*)rays_hit.normal.data());
 		} else {
 			// Store colors in the normal buffer
 			uint32_t n_elements = next_multiple(n_hit, BATCH_SIZE_MULTIPLE);
+6 −4		include/tiny-cuda-nn/common_device.h
+4 −3		include/tiny-cuda-nn/network.h
+2 −2		include/tiny-cuda-nn/network_with_input_encoding.h
+2 −2		include/tiny-cuda-nn/networks/cutlass_mlp.h
+2 −2		include/tiny-cuda-nn/networks/cutlass_resnet.h
+2 −2		include/tiny-cuda-nn/networks/fully_fused_mlp.h
+3 −3		src/network.cu