Star32
8
代码
代码
Issues8
Pull Requests2
Actions
项目讨论
Wiki
分析
Star32
8
已合并
整改维度信息 #13
xbyb26创建于 19 天前
整改维度信息 #13
已合并
xbyb26创建于 19 天前
从xbyb26/gauss-splatmaster合入到cann/gauss-splatmaster
共 12 个文件变更+156-156
Mdocs/en/gauss_splat_api_en.md+57-57
@@ -38,12 +38,12 @@ Renders 3D Gaussian point clouds to 2D images.
38 38 
39**Return Values**:39**Return Values**:
40 40 
41-- `render_colors` (Tensor): Rendered color image, shape (C, 3, H, W)41+- `render_colors` (Tensor): Rendered color image, shape (C, H, W, 3)
42-- `render_depth` (Tensor): Rendered depth image, shape (C, 1, H, W)42+- `render_depth` (Tensor): Rendered depth image, shape (C, H, W, 1)
43- `info` (dict): Metadata dictionary, containing:43- `info` (dict): Metadata dictionary, containing:
44 - `gaussian_ids`: Gaussian point IDs (currently None)44 - `gaussian_ids`: Gaussian point IDs (currently None)
45- - `means2d`: 2D projection coordinates45+ - `means2d`: 2D projection coordinates, shape (B, C, 2, N)
46- - `radii`: Projection radii46+ - `radii`: Projection radii, shape (B, C, 2, N)
47 - `width`: Image width47 - `width`: Image width
48 - `height`: Image height48 - `height`: Image height
49 - `n_cameras`: Number of cameras49 - `n_cameras`: Number of cameras
@@ -73,7 +73,7 @@ render_colors, render_depth, info = rasterizer.rasterization(
73**Parameters**:73**Parameters**:
74 74 
75- `means` (Tensor): Gaussian center positions, shape (B, N, 3)75- `means` (Tensor): Gaussian center positions, shape (B, N, 3)
76-- `colors` (Tensor): Color values, shape (B, N, 3) or (B, C, N, 3)76+- `colors` (Tensor): Color values, shape (B, 3, N)
77- `covars` (Tensor, optional): Covariance matrix, shape (B, N, 3, 3). Mutually exclusive with quat/scales77- `covars` (Tensor, optional): Covariance matrix, shape (B, N, 3, 3). Mutually exclusive with quat/scales
78- `quat` (Tensor, optional): Quaternions for rotation, shape (B, N, 4). Mutually exclusive with covars78- `quat` (Tensor, optional): Quaternions for rotation, shape (B, N, 4). Mutually exclusive with covars
79- `scales` (Tensor, optional): Scale parameters, shape (B, N, 3). Used together with quat79- `scales` (Tensor, optional): Scale parameters, shape (B, N, 3). Used together with quat
@@ -90,14 +90,14 @@ render_colors, render_depth, info = rasterizer.rasterization(
90 90 
91**Return Values**:91**Return Values**:
92 92 
93-- `means2d` (Tensor): 2D projection coordinates, shape (B, C, N, 2)93+- `means2d` (Tensor): 2D projection coordinates, shape (B, C, 2, N)
94- `depths` (Tensor): Depth values, shape (B, C, N)94- `depths` (Tensor): Depth values, shape (B, C, N)
95-- `conics` (Tensor): 2D covariance inverse matrix (conic parameters), shape (B, C, N, 3)95+- `conics` (Tensor): 2D covariance inverse matrix (conic parameters), shape (B, C, 3, N)
96- `opacities` (Tensor): Filtered opacities, shape (B, C, N)96- `opacities` (Tensor): Filtered opacities, shape (B, C, N)
97-- `radius` (Tensor): Projection radii, shape (B, C, N)97+- `radius` (Tensor): Projection radii, shape (B, C, 2, N)
98-- `covars2d` (Tensor): 2D covariance matrices, shape (B, C, N, 2, 2)98+- `covars2d` (Tensor): 2D covariance matrices, shape (B, C, 3, N)
99-- `colors` (Tensor): Filtered colors, shape (B, C, N, 3)99+- `colors` (Tensor): Filtered colors, shape (B, C, 3, N)
100-- `cnt` (Tensor): Number of valid Gaussian points100+- `cnt` (Tensor): Number of valid Gaussian points, shape (B, C)
101 101 
102**Example**:102**Example**:
103 103 
@@ -133,17 +133,17 @@ means2d, depths, conics, opacities, radius, covars2d, colors, cnt = \
133- `opacities` (Tensor): Opacities, shape (1, N)133- `opacities` (Tensor): Opacities, shape (1, N)
134- `colors` (Tensor): Colors, shape (3, N)134- `colors` (Tensor): Colors, shape (3, N)
135- `depths` (Tensor, optional): Depth values, shape (1, N). If None, depth is not rendered135- `depths` (Tensor, optional): Depth values, shape (1, N). If None, depth is not rendered
136-- `tile_coords` (Tensor): Tile coordinates136+- `tile_coords` (Tensor): Tile coordinates, shape (tileNum, 2, nPixel)
137-- `offsets` (Tensor): Offsets137+- `offsets` (Tensor): Offsets, shape(vectorCnt + (TileNum * 2))
138-- `sorted_gs_ids` (Tensor): Sorted Gaussian point IDs138+- `sorted_gs_ids` (Tensor): Sorted Gaussian point IDs, shape (totalGauss)
139 139 
140**Return Values**:140**Return Values**:
141 141 
142- If `depths` is provided:142- If `depths` is provided:
143- - `color` (Tensor): Rendered color image143+ - `color` (Tensor): Rendered color image, shape (3, tileNum, nPixel)
144- - `depth` (Tensor): Rendered depth image144+ - `depth` (Tensor): Rendered depth image, shape (1, tileNum, nPixel)
145- If `depths` is not provided:145- If `depths` is not provided:
146- - `color` (Tensor): Rendered color image146+ - `color` (Tensor): Rendered color image, shape (3, tileNum, nPixel)
147 147 
148**Example**:148**Example**:
149 149 
@@ -178,7 +178,7 @@ render_colors, render_depths = gauss_splat.calc_render(
178 178 
179**Return Values**:179**Return Values**:
180 180 
181-- `output` (Tensor): Computed color values, shape (B, N, 3)181+- `output` (Tensor): Computed color values, shape (B, 3, N)
182 182 
183**Example**:183**Example**:
184 184 
@@ -200,16 +200,16 @@ colors = gauss_splat.spherical_harmonics(
200 200 
201**Parameters**:201**Parameters**:
202 202 
203-- `lb_sched` (Tensor): Load balancing scheduling tensor203+- `lb_sched` (Tensor): Load balancing scheduling tensor, shape(B, C, schedule_num)
204-- `gaussian_cnt` (Tensor): Gaussian point count per tile204+- `gaussian_cnt` (Tensor): Gaussian point count per tile, shape(B, C, tile_num, 1)
205-- `depths` (Tensor): Depth values205+- `depths` (Tensor): Depth values, shape(B, C, tile_num, N)
206-- `gs_ids` (Tensor): Gaussian point IDs206+- `gs_ids` (Tensor): Gaussian point IDs, shape(B, C, tile_num, N)
207-- `sorted_offset` (Tensor): Sorting offset207+- `sorted_offset` (Tensor): Sorting offset, shape(B*C)
208- `max_tile_gauss` (int): Maximum number of Gaussian points per tile208- `max_tile_gauss` (int): Maximum number of Gaussian points per tile
209 209 
210**Return Values**:210**Return Values**:
211 211 
212-- `sorted_gs_ids` (Tensor): Sorted Gaussian point IDs212+- `sorted_gs_ids` (Tensor): Sorted Gaussian point IDs, shape(totalGauss)
213 213 
214**Example**:214**Example**:
215 215 
@@ -234,11 +234,11 @@ sorted_gs_ids = gauss_splat.gaussian_sort(
234 234 
235**Parameters**:235**Parameters**:
236 236 
237-- `means2d` (Tensor): 2D projection coordinates, shape (B, C, N, 2)237+- `means2d` (Tensor): 2D projection coordinates, shape (B, C, 2, N)
238-- `opacity` (Tensor): Opacities, shape (B, C, N)238+- `opacity` (Tensor): Opacities, shape (B, C, 1, N)
239-- `conics` (Tensor): Covariance inverse matrix, shape (B, C, N, 3)239+- `conics` (Tensor): Covariance inverse matrix, shape(B, C, 3, N)
240-- `covars2d` (Tensor): 2D covariance matrix, shape (B, C, N, 2, 2)240+- `covars2d` (Tensor): 2D covariance matrix, shape(B, C, 3, N)
241-- `depths` (Tensor): Depth values, shape (B, C, N)241+- `depths` (Tensor): Depth values, shape (B, C, 1, N)
242- `cnt` (Tensor): Valid Gaussian point count, shape (B, C)242- `cnt` (Tensor): Valid Gaussian point count, shape (B, C)
243- `tile_grid` (Tensor): Tile grid coordinates243- `tile_grid` (Tensor): Tile grid coordinates
244- `image_width` (int): Image width244- `image_width` (int): Image width
@@ -247,10 +247,10 @@ sorted_gs_ids = gauss_splat.gaussian_sort(
247 247 
248**Return Values**:248**Return Values**:
249 249 
250-- `tile_sum` (Tensor): Gaussian point sum per tile250+- `tile_sum` (Tensor): Gaussian point sum per tile, shape (B, C, tile_num, 1)
251-- `tile_offset` (Tensor): Tile offsets251+- `tile_offset` (Tensor): Tile offsets, shape (B, C, tile_num, 1)
252-- `tile_depths` (Tensor): Tile depths252+- `tile_depths` (Tensor): Tile depths, shape (B, C, tile_num, N)
253-- `gauss_index` (Tensor): Gaussian point indices253+- `gauss_index` (Tensor): Gaussian point indices, shape (B, C, tile_num, N)
254 254 
255**Example**:255**Example**:
256 256 
@@ -280,16 +280,16 @@ tile_sums, tile_offsets, tile_depths, tile_gauss_ids = \
280 280 
281**Parameters**:281**Parameters**:
282 282 
283-- `means` (Tensor): 3D Gaussian center positions283+- `means` (Tensor): 3D Gaussian center positions, shape(B, 3, N)
284-- `colors` (Tensor): Color values284+- `colors` (Tensor): Color values, shape(B, 3, N)
285-- `det` (Tensor): Covariance determinant285+- `det` (Tensor): Covariance determinant, shape(B, C, N)
286-- `opacities` (Tensor): Opacities286+- `opacities` (Tensor): Opacities, shape(B, N)
287-- `means2d` (Tensor): 2D projection coordinates287+- `means2d` (Tensor): 2D projection coordinates, shape(B, C, 2, N)
288-- `depths` (Tensor): Depth values288+- `depths` (Tensor): Depth values, shape(B, C, N)
289-- `radius` (Tensor): Projection radii289+- `radius` (Tensor): Projection radii, shape(B, C, 2, N)
290-- `conics` (Tensor): Covariance inverse matrix290+- `conics` (Tensor): Covariance inverse matrix, shape(B, C, 3, N)
291-- `covars2d` (Tensor): 2D covariance matrix291+- `covars2d` (Tensor): 2D covariance matrix, shape(B, C, 3, N)
292-- `compensations` (Tensor, optional): Compensation factors292+- `compensations` (Tensor, optional): Compensation factors, shape(B, C, N)
293- `width` (int): Image width293- `width` (int): Image width
294- `height` (int): Image height294- `height` (int): Image height
295- `near_plane` (float): Near plane distance295- `near_plane` (float): Near plane distance
@@ -297,16 +297,16 @@ tile_sums, tile_offsets, tile_depths, tile_gauss_ids = \
297 297 
298**Return Values**:298**Return Values**:
299 299 
300-- `means_culling`: Filtered 3D coordinates300+- `means_culling`: Filtered 3D coordinates, shape(B, C, 3, N)
301-- `colors_culling`: Filtered colors301+- `colors_culling`: Filtered colors, shape(B, C, 3, N)
302-- `means2d_culling`: Filtered 2D coordinates302+- `means2d_culling`: Filtered 2D coordinates, shape(B, C, 2, N)
303-- `depths_culling`: Filtered depths303+- `depths_culling`: Filtered depths, shape(B, C, N)
304-- `radius_culling`: Filtered radii304+- `radius_culling`: Filtered radii, shape(B, C, 2, N)
305-- `covars2d_culling`: Filtered 2D covariances305+- `covars2d_culling`: Filtered 2D covariances, shape(B, C, 3, N)
306-- `conics_culling`: Filtered covariance inverse matrices306+- `conics_culling`: Filtered covariance inverse matrices, shape(B, C, 3, N)
307-- `opacities_culling`: Filtered opacities307+- `opacities_culling`: Filtered opacities, shape(B, C, N)
308-- `proj_filter`: Projection filter308+- `proj_filter`: Projection filter, shape(B, C, ceil(N/8))
309-- `cnt`: Number of valid points309+- `cnt`: Number of valid points, shape(B, C)
310 310 
311---311---
312 312 
@@ -318,12 +318,12 @@ tile_sums, tile_offsets, tile_depths, tile_gauss_ids = \
318 318 
319**Parameters**:319**Parameters**:
320 320 
321-- `nums_tensor` (Tensor): Number of Gaussian points per tile321+- `nums_tensor` (Tensor): Number of Gaussian points per tile, shape(B, C, T)
322- `num_bins` (int): Number of bins (number of vector processors)322- `num_bins` (int): Number of bins (number of vector processors)
323 323 
324**Return Values**:324**Return Values**:
325 325 
326-- `lb_sched_tensor` (Tensor): Load balancing scheduling tensor326+- `lb_sched_tensor` (Tensor): Load balancing scheduling tensor, shape(B, C, M)
327 327 
328**Example**:328**Example**:
329 329 
@@ -408,8 +408,8 @@ render_colors, render_depths, info = rasterizer.rasterization(
408 camera_model="pinhole"408 camera_model="pinhole"
409)409)
410 410 
411-print(f"Rendered color shape: {render_colors.shape}") # (1, 3, 1080, 1920)411+print(f"Rendered color shape: {render_colors.shape}") # (1, 1080, 1920, 3)
412-print(f"Rendered depth shape: {render_depths.shape}") # (1, 1, 1080, 1920)412+print(f"Rendered depth shape: {render_depths.shape}") # (1, 1080, 1920, 1)
413```413```
414 414 
415---415---
Mdocs/en/kernels/aclnnCalcRenderBwdVarClipGsids_en.md+2-2
@@ -148,7 +148,7 @@ aclnnStatus aclnnCalcRenderBwdVarClipGsids(
148 <td>Empty tensors are not supported.</td>148 <td>Empty tensors are not supported.</td>
149 <td>INT64</td>149 <td>INT64</td>
150 <td>ND</td>150 <td>ND</td>
151- <td>(totalGauss)</td>151+ <td>(tileNum + totalGauss)</td>
152 <td>Yes</td>152 <td>Yes</td>
153 </tr>153 </tr>
154 <tr>154 <tr>
@@ -158,7 +158,7 @@ aclnnStatus aclnnCalcRenderBwdVarClipGsids(
158 <td>Empty tensors are not supported.</td>158 <td>Empty tensors are not supported.</td>
159 <td>UINT8</td>159 <td>UINT8</td>
160 <td>ND</td>160 <td>ND</td>
161- <td>(nPixel)</td>161+ <td>(totalGauss, 2)</td>
162 <td>Yes</td>162 <td>Yes</td>
163 </tr>163 </tr>
164 <tr>164 <tr>
Mdocs/en/kernels/aclnnFullyFusedProjectionBwd_en.md+11-11
@@ -135,7 +135,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
135 <td>Empty tensors are not supported.</td>135 <td>Empty tensors are not supported.</td>
136 <td>FLOAT</td>136 <td>FLOAT</td>
137 <td>ND</td>137 <td>ND</td>
138- <td>(camera_num, 4, 4)</td>138+ <td>(batch_size, camera_num, 4, 4)</td>
139 <td>Yes</td>139 <td>Yes</td>
140 </tr>140 </tr>
141 <tr>141 <tr>
@@ -145,7 +145,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
145 <td>Empty tensors are not supported.</td>145 <td>Empty tensors are not supported.</td>
146 <td>FLOAT</td>146 <td>FLOAT</td>
147 <td>ND</td>147 <td>ND</td>
148- <td>(camera_num, 3, 3)</td>148+ <td>(batch_size, camera_num, 3, 3)</td>
149 <td>Yes</td>149 <td>Yes</td>
150 </tr>150 </tr>
151 <tr>151 <tr>
@@ -165,7 +165,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
165 <td>Empty tensors are not supported.</td>165 <td>Empty tensors are not supported.</td>
166 <td>FLOAT</td>166 <td>FLOAT</td>
167 <td>ND</td>167 <td>ND</td>
168- <td>(batch_size, camera_num, 1, gaussian_num)</td>168+ <td>(batch_size, camera_num, gaussian_num)</td>
169 <td>Yes</td>169 <td>Yes</td>
170 </tr>170 </tr>
171 <tr>171 <tr>
@@ -185,7 +185,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
185 <td>Empty tensors are not supported.</td>185 <td>Empty tensors are not supported.</td>
186 <td>FLOAT</td>186 <td>FLOAT</td>
187 <td>ND</td>187 <td>ND</td>
188- <td>(batch_size, 3, gaussian_num)</td>188+ <td>(batch_size, camera_num, 3, gaussian_num)</td>
189 <td>Yes</td>189 <td>Yes</td>
190 </tr>190 </tr>
191 <tr>191 <tr>
@@ -195,7 +195,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
195 <td>Empty tensors are not supported.</td>195 <td>Empty tensors are not supported.</td>
196 <td>FLOAT</td>196 <td>FLOAT</td>
197 <td>ND</td>197 <td>ND</td>
198- <td>(batch_size, gaussian_num)</td>198+ <td>(batch_size, camera_num, gaussian_num)</td>
199 <td>Yes</td>199 <td>Yes</td>
200 </tr>200 </tr>
201 <tr>201 <tr>
@@ -205,7 +205,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
205 <td>Empty tensors are not supported.</td>205 <td>Empty tensors are not supported.</td>
206 <td>UINT8</td>206 <td>UINT8</td>
207 <td>ND</td>207 <td>ND</td>
208- <td>(batch_size, camera_num, gaussian_num)</td>208+ <td>(batch_size, camera_num, ceil(gaussian_num/8))</td>
209 <td>Yes</td>209 <td>Yes</td>
210 </tr>210 </tr>
211 <tr>211 <tr>
@@ -215,7 +215,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
215 <td>Optional input, can pass nullptr.</td>215 <td>Optional input, can pass nullptr.</td>
216 <td>FLOAT</td>216 <td>FLOAT</td>
217 <td>ND</td>217 <td>ND</td>
218- <td>(batch_size, camera_num, 1, gaussian_num)</td>218+ <td>(batch_size, camera_num, gaussian_num)</td>
219 <td>Yes</td>219 <td>Yes</td>
220 </tr>220 </tr>
221 <tr>221 <tr>
@@ -245,7 +245,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
245 <td>Empty tensors are not supported.</td>245 <td>Empty tensors are not supported.</td>
246 <td>FLOAT</td>246 <td>FLOAT</td>
247 <td>ND</td>247 <td>ND</td>
248- <td>(batch_size, 3, gaussian_num)</td>248+ <td>(batch_size, gaussian_num, 3)</td>
249 <td>Yes</td>249 <td>Yes</td>
250 </tr>250 </tr>
251 <tr>251 <tr>
@@ -255,7 +255,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
255 <td>Empty tensors are not supported.</td>255 <td>Empty tensors are not supported.</td>
256 <td>FLOAT</td>256 <td>FLOAT</td>
257 <td>ND</td>257 <td>ND</td>
258- <td>(batch_size, 4, gaussian_num)</td>258+ <td>(batch_size, gaussian_num, 4)</td>
259 <td>Yes</td>259 <td>Yes</td>
260 </tr>260 </tr>
261 <tr>261 <tr>
@@ -265,7 +265,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
265 <td>Empty tensors are not supported.</td>265 <td>Empty tensors are not supported.</td>
266 <td>FLOAT</td>266 <td>FLOAT</td>
267 <td>ND</td>267 <td>ND</td>
268- <td>(batch_size, 3, gaussian_num)</td>268+ <td>(batch_size, gaussian_num, 3)</td>
269 <td>Yes</td>269 <td>Yes</td>
270 </tr>270 </tr>
271 <tr>271 <tr>
@@ -275,7 +275,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
275 <td>Empty tensors are not supported.</td>275 <td>Empty tensors are not supported.</td>
276 <td>FLOAT</td>276 <td>FLOAT</td>
277 <td>ND</td>277 <td>ND</td>
278- <td>(batch_size, 3, 3, gaussian_num)</td>278+ <td>(batch_size, camera_num, 3, 3)</td>
279 <td>Yes</td>279 <td>Yes</td>
280 </tr>280 </tr>
281 <tr>281 <tr>
Mdocs/en/kernels/aclnnGaussianFilter_en.md+4-4
@@ -142,7 +142,7 @@ aclnnStatus aclnnGaussianFilter(
142 <td>Empty tensors are not supported.</td>142 <td>Empty tensors are not supported.</td>
143 <td>FLOAT</td>143 <td>FLOAT</td>
144 <td>ND</td>144 <td>ND</td>
145- <td>(batch_size, camera_num, 1, gaussian_num)</td>145+ <td>(batch_size, camera_num, gaussian_num)</td>
146 <td>Yes</td>146 <td>Yes</td>
147 </tr>147 </tr>
148 <tr>148 <tr>
@@ -182,7 +182,7 @@ aclnnStatus aclnnGaussianFilter(
182 <td>Optional input, can pass nullptr.</td>182 <td>Optional input, can pass nullptr.</td>
183 <td>FLOAT</td>183 <td>FLOAT</td>
184 <td>ND</td>184 <td>ND</td>
185- <td>(batch_size, camera_num, 1, gaussian_num)</td>185+ <td>(batch_size, camera_num, gaussian_num)</td>
186 <td>Yes</td>186 <td>Yes</td>
187 </tr>187 </tr>
188 <tr>188 <tr>
@@ -262,7 +262,7 @@ aclnnStatus aclnnGaussianFilter(
262 <td>Empty tensors are not supported.</td>262 <td>Empty tensors are not supported.</td>
263 <td>FLOAT</td>263 <td>FLOAT</td>
264 <td>ND</td>264 <td>ND</td>
265- <td>(batch_size, camera_num, 1, gaussian_num)</td>265+ <td>(batch_size, camera_num, gaussian_num)</td>
266 <td>Yes</td>266 <td>Yes</td>
267 </tr>267 </tr>
268 <tr>268 <tr>
@@ -312,7 +312,7 @@ aclnnStatus aclnnGaussianFilter(
312 <td>Empty tensors are not supported.</td>312 <td>Empty tensors are not supported.</td>
313 <td>UINT8</td>313 <td>UINT8</td>
314 <td>ND</td>314 <td>ND</td>
315- <td>(batch_size, camera_num, gaussian_num)</td>315+ <td>(batch_size, camera_num, ceil(gaussian_num/8))</td>
316 <td>Yes</td>316 <td>Yes</td>
317 </tr>317 </tr>
318 <tr>318 <tr>
Mdocs/en/kernels/aclnnGaussianSort_en.md+2-2
@@ -115,7 +115,7 @@ aclnnStatus aclnnGaussianSort(
115 <td>Empty tensors are not supported.</td>115 <td>Empty tensors are not supported.</td>
116 <td>INT64</td>116 <td>INT64</td>
117 <td>ND</td>117 <td>ND</td>
118- <td>(batch_size, camera_num, ...)</td>118+ <td>(batch_size * camera_num)</td>
119 <td>Yes</td>119 <td>Yes</td>
120 </tr>120 </tr>
121 <tr>121 <tr>
@@ -135,7 +135,7 @@ aclnnStatus aclnnGaussianSort(
135 <td>Empty tensors are not supported.</td>135 <td>Empty tensors are not supported.</td>
136 <td>INT32</td>136 <td>INT32</td>
137 <td>ND</td>137 <td>ND</td>
138- <td>(batch_size, camera_num, ...)</td>138+ <td>(totalGauss)</td>
139 <td>Yes</td>139 <td>Yes</td>
140 </tr>140 </tr>
141 <tr>141 <tr>
Mdocs/en/kernels/aclnnSphericalHarmonicsBwd_en.md+2-2
@@ -101,7 +101,7 @@ aclnnStatus aclnnSphericalHarmonicsBwd(
101 <td>degree</td>101 <td>degree</td>
102 <td>Input</td>102 <td>Input</td>
103 <td>Spherical harmonics degree used.</td>103 <td>Spherical harmonics degree used.</td>
104- <td>Supports 0 to 3.</td>104+ <td>Supports 0 to 4.</td>
105 <td>INT64</td>105 <td>INT64</td>
106 <td>-</td>106 <td>-</td>
107 <td>-</td>107 <td>-</td>
@@ -170,7 +170,7 @@ aclnnStatus aclnnSphericalHarmonicsBwd(
170 <tr>170 <tr>
171 <td>ACLNN_ERR_PARAM_INVALID</td>171 <td>ACLNN_ERR_PARAM_INVALID</td>
172 <td>161002</td>172 <td>161002</td>
173- <td>The data types and formats of dirs, coeffs, vColors, vDirs, and vCoeffs are not within the supported range, or degree is not within the range of 0 to 3.</td>173+ <td>The data types and formats of dirs, coeffs, vColors, vDirs, and vCoeffs are not within the supported range, or degree is not within the range of 0 to 4.</td>
174 </tr>174 </tr>
175 </tbody>175 </tbody>
176 </table>176 </table>
Mdocs/zh/gauss_splat_api.md+57-57
@@ -38,12 +38,12 @@
38 38 
39**返回值**:39**返回值**:
40 40 
41-- `render_colors` (Tensor): 渲染的颜色图像，形状 (C, 3, H, W)41+- `render_colors` (Tensor): 渲染的颜色图像，形状 (C, H, W, 3)
42-- `render_depth` (Tensor): 渲染的深度图像，形状 (C, 1, H, W)42+- `render_depth` (Tensor): 渲染的深度图像，形状 (C, H, W, 1)
43- `info` (dict): 元数据字典，包含:43- `info` (dict): 元数据字典，包含:
44 - `gaussian_ids`: 高斯点ID（当前为None）44 - `gaussian_ids`: 高斯点ID（当前为None）
45- - `means2d`: 2D投影坐标45+ - `means2d`: 2D投影坐标，形状 (B, C, 2, N)
46- - `radii`: 投影半径46+ - `radii`: 投影半径，形状 (B, C, 2, N)
47 - `width`: 图像宽度47 - `width`: 图像宽度
48 - `height`: 图像高度48 - `height`: 图像高度
49 - `n_cameras`: 相机数量49 - `n_cameras`: 相机数量
@@ -73,7 +73,7 @@ render_colors, render_depth, info = rasterizer.rasterization(
73**参数**:73**参数**:
74 74 
75- `means` (Tensor): 高斯中心位置，形状 (B, N, 3)75- `means` (Tensor): 高斯中心位置，形状 (B, N, 3)
76-- `colors` (Tensor): 颜色值，形状 (B, N, 3) 或 (B, C, N, 3)76+- `colors` (Tensor): 颜色值，形状 (B, 3, N)
77- `covars` (Tensor, optional): 协方差矩阵，形状 (B, N, 3, 3)。与quat/scales互斥77- `covars` (Tensor, optional): 协方差矩阵，形状 (B, N, 3, 3)。与quat/scales互斥
78- `quat` (Tensor, optional): 四元数表示旋转，形状 (B, N, 4)。与covars互斥78- `quat` (Tensor, optional): 四元数表示旋转，形状 (B, N, 4)。与covars互斥
79- `scales` (Tensor, optional): 缩放参数，形状 (B, N, 3)。与quat一起使用79- `scales` (Tensor, optional): 缩放参数，形状 (B, N, 3)。与quat一起使用
@@ -90,14 +90,14 @@ render_colors, render_depth, info = rasterizer.rasterization(
90 90 
91**返回值**:91**返回值**:
92 92 
93-- `means2d` (Tensor): 2D投影坐标，形状 (B, C, N, 2)93+- `means2d` (Tensor): 2D投影坐标，形状 (B, C, 2, N)
94- `depths` (Tensor): 深度值，形状 (B, C, N)94- `depths` (Tensor): 深度值，形状 (B, C, N)
95-- `conics` (Tensor): 2D协方差逆矩阵（锥形参数），形状 (B, C, N, 3)95+- `conics` (Tensor): 2D协方差逆矩阵（锥形参数），形状 (B, C, 3, N)
96- `opacities` (Tensor): 过滤后的不透明度，形状 (B, C, N)96- `opacities` (Tensor): 过滤后的不透明度，形状 (B, C, N)
97-- `radius` (Tensor): 投影半径，形状 (B, C, N)97+- `radius` (Tensor): 投影半径，形状 (B, C, 2, N)
98-- `covars2d` (Tensor): 2D协方差矩阵，形状 (B, C, N, 2, 2)98+- `covars2d` (Tensor): 2D协方差矩阵，形状 (B, C, 3, N)
99-- `colors` (Tensor): 过滤后的颜色，形状 (B, C, N, 3)99+- `colors` (Tensor): 过滤后的颜色，形状 (B, C, 3, N)
100-- `cnt` (Tensor): 有效高斯点数量100+- `cnt` (Tensor): 有效高斯点数量, 形状 (B, C)
101 101 
102**示例**:102**示例**:
103 103 
@@ -133,17 +133,17 @@ means2d, depths, conics, opacities, radius, covars2d, colors, cnt = \
133- `opacities` (Tensor): 不透明度，形状 (1, N)133- `opacities` (Tensor): 不透明度，形状 (1, N)
134- `colors` (Tensor): 颜色，形状 (3, N)134- `colors` (Tensor): 颜色，形状 (3, N)
135- `depths` (Tensor, optional): 深度值，形状 (1, N)。如为None则不渲染深度135- `depths` (Tensor, optional): 深度值，形状 (1, N)。如为None则不渲染深度
136-- `tile_coords` (Tensor): 分块坐标136+- `tile_coords` (Tensor): 分块坐标，形状 (tileNum, 2, nPixel)
137-- `offsets` (Tensor): 偏移量137+- `offsets` (Tensor): 偏移量，形状 (vectorCnt + (TileNum * 2))
138-- `sorted_gs_ids` (Tensor): 排序后的高斯点ID138+- `sorted_gs_ids` (Tensor): 排序后的高斯点ID，形状 (totalGauss)
139 139 
140**返回值**:140**返回值**:
141 141 
142- 如果提供`depths`:142- 如果提供`depths`:
143- - `color` (Tensor): 渲染的颜色图像143+ - `color` (Tensor): 渲染的颜色图像，形状 (3, tileNum, nPixel)
144- - `depth` (Tensor): 渲染的深度图像144+ - `depth` (Tensor): 渲染的深度图像，形状 (1, tileNum, nPixel)
145- 如果不提供`depths`:145- 如果不提供`depths`:
146- - `color` (Tensor): 渲染的颜色图像146+ - `color` (Tensor): 渲染的颜色图像，形状 (3, tileNum, nPixel)
147 147 
148**示例**:148**示例**:
149 149 
@@ -178,7 +178,7 @@ render_colors, render_depths = gauss_splat.calc_render(
178 178 
179**返回值**:179**返回值**:
180 180 
181-- `output` (Tensor): 计算得到的颜色值，形状 (B, N, 3)181+- `output` (Tensor): 计算得到的颜色值，形状 (B, 3, N)
182 182 
183**示例**:183**示例**:
184 184 
@@ -200,16 +200,16 @@ colors = gauss_splat.spherical_harmonics(
200 200 
201**参数**:201**参数**:
202 202 
203-- `lb_sched` (Tensor): 负载均衡调度张量203+- `lb_sched` (Tensor): 负载均衡调度张量，形状 (B, C, schedule_num)
204-- `gaussian_cnt` (Tensor): 每个tile的高斯点计数204+- `gaussian_cnt` (Tensor): 每个tile的高斯点计数，形状 (B, C, tile_num, 1)
205-- `depths` (Tensor): 深度值205+- `depths` (Tensor): 深度值，形状 (B, C, tile_num, N)
206-- `gs_ids` (Tensor): 高斯点ID206+- `gs_ids` (Tensor): 高斯点ID，形状 (B, C, tile_num, N)
207-- `sorted_offset` (Tensor): 排序偏移量207+- `sorted_offset` (Tensor): 排序偏移量，形状 (B*C)
208- `max_tile_gauss` (int): 单个tile最大高斯点数208- `max_tile_gauss` (int): 单个tile最大高斯点数
209 209 
210**返回值**:210**返回值**:
211 211 
212-- `sorted_gs_ids` (Tensor): 排序后的高斯点ID212+- `sorted_gs_ids` (Tensor): 排序后的高斯点ID，一维展平张量，形状 (totalGauss)
213 213 
214**示例**:214**示例**:
215 215 
@@ -234,11 +234,11 @@ sorted_gs_ids = gauss_splat.gaussian_sort(
234 234 
235**参数**:235**参数**:
236 236 
237-- `means2d` (Tensor): 2D投影坐标，形状 (B, C, N, 2)237+- `means2d` (Tensor): 2D投影坐标，形状 (B, C, 2, N)
238-- `opacity` (Tensor): 不透明度，形状 (B, C, N)238+- `opacity` (Tensor): 不透明度，形状 (B, C, 1, N)
239-- `conics` (Tensor): 协方差逆矩阵，形状 (B, C, N, 3)239+- `conics` (Tensor): 协方差逆矩阵，形状 (B, C, 3, N)
240-- `covars2d` (Tensor): 2D协方差矩阵，形状 (B, C, N, 2, 2)240+- `covars2d` (Tensor): 2D协方差矩阵，形状 (B, C, 3, N)
241-- `depths` (Tensor): 深度值，形状 (B, C, N)241+- `depths` (Tensor): 深度值，形状 (B, C, 1, N)
242- `cnt` (Tensor): 有效高斯点计数，形状 (B, C)242- `cnt` (Tensor): 有效高斯点计数，形状 (B, C)
243- `tile_grid` (Tensor): tile网格坐标243- `tile_grid` (Tensor): tile网格坐标
244- `image_width` (int): 图像宽度244- `image_width` (int): 图像宽度
@@ -247,10 +247,10 @@ sorted_gs_ids = gauss_splat.gaussian_sort(
247 247 
248**返回值**:248**返回值**:
249 249 
250-- `tile_sum` (Tensor): 每个tile的高斯点和250+- `tile_sum` (Tensor): 每个tile的高斯点和，形状(B, C, tile_num, 1)
251-- `tile_offset` (Tensor): tile偏移量251+- `tile_offset` (Tensor): tile偏移量，形状(B, C, tile_num, 1)
252-- `tile_depths` (Tensor): tile深度252+- `tile_depths` (Tensor): tile深度，形状(B, C, tile_num, N)
253-- `gauss_index` (Tensor): 高斯点索引253+- `gauss_index` (Tensor): 高斯点索引，形状(B, C, tile_num, N)
254 254 
255**示例**:255**示例**:
256 256 
@@ -280,16 +280,16 @@ tile_sums, tile_offsets, tile_depths, tile_gauss_ids = \
280 280 
281**参数**:281**参数**:
282 282 
283-- `means` (Tensor): 3D高斯中心位置283+- `means` (Tensor): 3D高斯中心位置，形状 (B, 3, N)
284-- `colors` (Tensor): 颜色值284+- `colors` (Tensor): 颜色值，形状 (B, 3, N)
285-- `det` (Tensor): 协方差行列式285+- `det` (Tensor): 协方差行列式，形状 (B, C, N)
286-- `opacities` (Tensor): 不透明度286+- `opacities` (Tensor): 不透明度，形状 (B, N)
287-- `means2d` (Tensor): 2D投影坐标287+- `means2d` (Tensor): 2D投影坐标，形状 (B, C, 2, N)
288-- `depths` (Tensor): 深度值288+- `depths` (Tensor): 深度值，形状 (B, C, N)
289-- `radius` (Tensor): 投影半径289+- `radius` (Tensor): 投影半径，形状 (B, C, 2, N)
290-- `conics` (Tensor): 协方差逆矩阵290+- `conics` (Tensor): 协方差逆矩阵，形状 (B, C, 3, N)
291-- `covars2d` (Tensor): 2D协方差矩阵291+- `covars2d` (Tensor): 2D协方差矩阵，形状 (B, C, 3, N)
292-- `compensations` (Tensor, optional): 补偿因子292+- `compensations` (Tensor, optional): 补偿因子，形状 (B, C, N)
293- `width` (int): 图像宽度293- `width` (int): 图像宽度
294- `height` (int): 图像高度294- `height` (int): 图像高度
295- `near_plane` (float): 近平面距离295- `near_plane` (float): 近平面距离
@@ -297,16 +297,16 @@ tile_sums, tile_offsets, tile_depths, tile_gauss_ids = \
297 297 
298**返回值**:298**返回值**:
299 299 
300-- `means_culling`: 过滤后的3D坐标300+- `means_culling`: 过滤后的3D坐标，形状 (B, C, 3, N)
301-- `colors_culling`: 过滤后的颜色301+- `colors_culling`: 过滤后的颜色，形状 (B, C, 3, N)
302-- `means2d_culling`: 过滤后的2D坐标302+- `means2d_culling`: 过滤后的2D坐标，形状 (B, C, 2, N)
303-- `depths_culling`: 过滤后的深度303+- `depths_culling`: 过滤后的深度，形状 (B, C, N)
304-- `radius_culling`: 过滤后的半径304+- `radius_culling`: 过滤后的半径，形状 (B, C, 2, N)
305-- `covars2d_culling`: 过滤后的2D协方差305+- `covars2d_culling`: 过滤后的2D协方差，形状 (B, C, 3, N)
306-- `conics_culling`: 过滤后的协方差逆矩阵306+- `conics_culling`: 过滤后的协方差逆矩阵，形状 (B, C, 3, N)
307-- `opacities_culling`: 过滤后的不透明度307+- `opacities_culling`: 过滤后的不透明度，形状 (B, C, N)
308-- `proj_filter`: 投影过滤器308+- `proj_filter`: 投影过滤器，形状 (B, C, ceil(N/8))
309-- `cnt`: 有效点数量309+- `cnt`: 有效点数量，形状 (B, C)
310 310 
311---311---
312 312 
@@ -318,12 +318,12 @@ tile_sums, tile_offsets, tile_depths, tile_gauss_ids = \
318 318 
319**参数**:319**参数**:
320 320 
321-- `nums_tensor` (Tensor): 每个tile的高斯点数量321+- `nums_tensor` (Tensor): 每个tile的高斯点数量，形状 (B, C, T)
322- `num_bins` (int): 分箱数量（向量处理器数量）322- `num_bins` (int): 分箱数量（向量处理器数量）
323 323 
324**返回值**:324**返回值**:
325 325 
326-- `lb_sched_tensor` (Tensor): 负载均衡调度张量326+- `lb_sched_tensor` (Tensor): 负载均衡调度张量，形状 (B, C, M)
327 327 
328**示例**:328**示例**:
329 329 
@@ -408,8 +408,8 @@ render_colors, render_depths, info = rasterizer.rasterization(
408 camera_model="pinhole"408 camera_model="pinhole"
409)409)
410 410 
411-print(f"渲染颜色形状: {render_colors.shape}") # (1, 3, 1080, 1920)411+print(f"渲染颜色形状: {render_colors.shape}") # (1, 1080, 1920, 3)
412-print(f"渲染深度形状: {render_depths.shape}") # (1, 1, 1080, 1920)412+print(f"渲染深度形状: {render_depths.shape}") # (1, 1080, 1920, 1)
413```413```
414 414 
415---415---
Mdocs/zh/kernels/aclnnCalcRenderBwdVarClipGsids.md+2-2
@@ -148,7 +148,7 @@ aclnnStatus aclnnCalcRenderBwdVarClipGsids(
148 <td>不支持空tensor。</td>148 <td>不支持空tensor。</td>
149 <td>INT64</td>149 <td>INT64</td>
150 <td>ND</td>150 <td>ND</td>
151- <td>(totalGauss)</td>151+ <td>(tileNum + totalGauss)</td>
152 <td>√</td>152 <td>√</td>
153 </tr>153 </tr>
154 <tr>154 <tr>
@@ -158,7 +158,7 @@ aclnnStatus aclnnCalcRenderBwdVarClipGsids(
158 <td>不支持空tensor。</td>158 <td>不支持空tensor。</td>
159 <td>UINT8</td>159 <td>UINT8</td>
160 <td>ND</td>160 <td>ND</td>
161- <td>(nPixel)</td>161+ <td>(totalGauss, 2)</td>
162 <td>√</td>162 <td>√</td>
163 </tr>163 </tr>
164 <tr>164 <tr>
Mdocs/zh/kernels/aclnnFullyFusedProjectionBwd.md+11-11
@@ -135,7 +135,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
135 <td>不支持空tensor。</td>135 <td>不支持空tensor。</td>
136 <td>FLOAT</td>136 <td>FLOAT</td>
137 <td>ND</td>137 <td>ND</td>
138- <td>(camera_num, 4, 4)</td>138+ <td>(batch_size, camera_num, 4, 4)</td>
139 <td>√</td>139 <td>√</td>
140 </tr>140 </tr>
141 <tr>141 <tr>
@@ -145,7 +145,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
145 <td>不支持空tensor。</td>145 <td>不支持空tensor。</td>
146 <td>FLOAT</td>146 <td>FLOAT</td>
147 <td>ND</td>147 <td>ND</td>
148- <td>(camera_num, 3, 3)</td>148+ <td>(batch_size, camera_num, 3, 3)</td>
149 <td>√</td>149 <td>√</td>
150 </tr>150 </tr>
151 <tr>151 <tr>
@@ -165,7 +165,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
165 <td>不支持空tensor。</td>165 <td>不支持空tensor。</td>
166 <td>FLOAT</td>166 <td>FLOAT</td>
167 <td>ND</td>167 <td>ND</td>
168- <td>(batch_size, camera_num, 1, gaussian_num)</td>168+ <td>(batch_size, camera_num, gaussian_num)</td>
169 <td>√</td>169 <td>√</td>
170 </tr>170 </tr>
171 <tr>171 <tr>
@@ -185,7 +185,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
185 <td>不支持空tensor。</td>185 <td>不支持空tensor。</td>
186 <td>FLOAT</td>186 <td>FLOAT</td>
187 <td>ND</td>187 <td>ND</td>
188- <td>(batch_size, 3, gaussian_num)</td>188+ <td>(batch_size, camera_num, 3, gaussian_num)</td>
189 <td>√</td>189 <td>√</td>
190 </tr>190 </tr>
191 <tr>191 <tr>
@@ -195,7 +195,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
195 <td>不支持空tensor。</td>195 <td>不支持空tensor。</td>
196 <td>FLOAT</td>196 <td>FLOAT</td>
197 <td>ND</td>197 <td>ND</td>
198- <td>(batch_size, gaussian_num)</td>198+ <td>(batch_size, camera_num, gaussian_num)</td>
199 <td>√</td>199 <td>√</td>
200 </tr>200 </tr>
201 <tr>201 <tr>
@@ -205,7 +205,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
205 <td>不支持空tensor。</td>205 <td>不支持空tensor。</td>
206 <td>UINT8</td>206 <td>UINT8</td>
207 <td>ND</td>207 <td>ND</td>
208- <td>(batch_size, camera_num, gaussian_num)</td>208+ <td>(batch_size, camera_num, ceil(gaussian_num/8))</td>
209 <td>√</td>209 <td>√</td>
210 </tr>210 </tr>
211 <tr>211 <tr>
@@ -215,7 +215,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
215 <td>可选输入，可传入nullptr。</td>215 <td>可选输入，可传入nullptr。</td>
216 <td>FLOAT</td>216 <td>FLOAT</td>
217 <td>ND</td>217 <td>ND</td>
218- <td>(batch_size, camera_num, 1, gaussian_num)</td>218+ <td>(batch_size, camera_num, gaussian_num)</td>
219 <td>√</td>219 <td>√</td>
220 </tr>220 </tr>
221 <tr>221 <tr>
@@ -245,7 +245,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
245 <td>不支持空tensor。</td>245 <td>不支持空tensor。</td>
246 <td>FLOAT</td>246 <td>FLOAT</td>
247 <td>ND</td>247 <td>ND</td>
248- <td>(batch_size, 3, gaussian_num)</td>248+ <td>(batch_size, gaussian_num, 3)</td>
249 <td>√</td>249 <td>√</td>
250 </tr>250 </tr>
251 <tr>251 <tr>
@@ -255,7 +255,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
255 <td>不支持空tensor。</td>255 <td>不支持空tensor。</td>
256 <td>FLOAT</td>256 <td>FLOAT</td>
257 <td>ND</td>257 <td>ND</td>
258- <td>(batch_size, 4, gaussian_num)</td>258+ <td>(batch_size, gaussian_num, 4)</td>
259 <td>√</td>259 <td>√</td>
260 </tr>260 </tr>
261 <tr>261 <tr>
@@ -265,7 +265,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
265 <td>不支持空tensor。</td>265 <td>不支持空tensor。</td>
266 <td>FLOAT</td>266 <td>FLOAT</td>
267 <td>ND</td>267 <td>ND</td>
268- <td>(batch_size, 3, gaussian_num)</td>268+ <td>(batch_size, gaussian_num, 3)</td>
269 <td>√</td>269 <td>√</td>
270 </tr>270 </tr>
271 <tr>271 <tr>
@@ -275,7 +275,7 @@ aclnnStatus aclnnFullyFusedProjectionBwd(
275 <td>不支持空tensor。</td>275 <td>不支持空tensor。</td>
276 <td>FLOAT</td>276 <td>FLOAT</td>
277 <td>ND</td>277 <td>ND</td>
278- <td>(batch_size, 3, 3, gaussian_num)</td>278+ <td>(batch_size, camera_num, 3, 3)</td>
279 <td>√</td>279 <td>√</td>
280 </tr>280 </tr>
281 <tr>281 <tr>
Mdocs/zh/kernels/aclnnGaussianFilter.md+4-4
@@ -142,7 +142,7 @@ aclnnStatus aclnnGaussianFilter(
142 <td>不支持空tensor。</td>142 <td>不支持空tensor。</td>
143 <td>FLOAT</td>143 <td>FLOAT</td>
144 <td>ND</td>144 <td>ND</td>
145- <td>(batch_size, camera_num, 1, gaussian_num)</td>145+ <td>(batch_size, camera_num, gaussian_num)</td>
146 <td>√</td>146 <td>√</td>
147 </tr>147 </tr>
148 <tr>148 <tr>
@@ -182,7 +182,7 @@ aclnnStatus aclnnGaussianFilter(
182 <td>可选输入，可传入nullptr。</td>182 <td>可选输入，可传入nullptr。</td>
183 <td>FLOAT</td>183 <td>FLOAT</td>
184 <td>ND</td>184 <td>ND</td>
185- <td>(batch_size, camera_num, 1, gaussian_num)</td>185+ <td>(batch_size, camera_num, gaussian_num)</td>
186 <td>√</td>186 <td>√</td>
187 </tr>187 </tr>
188 <tr>188 <tr>
@@ -262,7 +262,7 @@ aclnnStatus aclnnGaussianFilter(
262 <td>不支持空tensor。</td>262 <td>不支持空tensor。</td>
263 <td>FLOAT</td>263 <td>FLOAT</td>
264 <td>ND</td>264 <td>ND</td>
265- <td>(batch_size, camera_num, 1, gaussian_num)</td>265+ <td>(batch_size, camera_num, gaussian_num)</td>
266 <td>√</td>266 <td>√</td>
267 </tr>267 </tr>
268 <tr>268 <tr>
@@ -312,7 +312,7 @@ aclnnStatus aclnnGaussianFilter(
312 <td>不支持空tensor。</td>312 <td>不支持空tensor。</td>
313 <td>UINT8</td>313 <td>UINT8</td>
314 <td>ND</td>314 <td>ND</td>
315- <td>(batch_size, camera_num, gaussian_num)</td>315+ <td>(batch_size, camera_num, ceil(gaussian_num/8))</td>
316 <td>√</td>316 <td>√</td>
317 </tr>317 </tr>
318 <tr>318 <tr>
Mdocs/zh/kernels/aclnnGaussianSort.md+2-2
@@ -115,7 +115,7 @@ aclnnStatus aclnnGaussianSort(
115 <td>不支持空tensor。</td>115 <td>不支持空tensor。</td>
116 <td>INT64</td>116 <td>INT64</td>
117 <td>ND</td>117 <td>ND</td>
118- <td>(batch_size, camera_num, ...)</td>118+ <td>(batch_size * camera_num)</td>
119 <td>√</td>119 <td>√</td>
120 </tr>120 </tr>
121 <tr>121 <tr>
@@ -135,7 +135,7 @@ aclnnStatus aclnnGaussianSort(
135 <td>不支持空tensor。</td>135 <td>不支持空tensor。</td>
136 <td>INT32</td>136 <td>INT32</td>
137 <td>ND</td>137 <td>ND</td>
138- <td>(batch_size, camera_num, ...)</td>138+ <td>(totalGauss)</td>
139 <td>√</td>139 <td>√</td>
140 </tr>140 </tr>
141 <tr>141 <tr>
Mdocs/zh/kernels/aclnnSphericalHarmonicsBwd.md+2-2
@@ -101,7 +101,7 @@ aclnnStatus aclnnSphericalHarmonicsBwd(
101 <td>degree</td>101 <td>degree</td>
102 <td>输入</td>102 <td>输入</td>
103 <td>使用的球谐阶数。</td>103 <td>使用的球谐阶数。</td>
104- <td>支持0~3。</td>104+ <td>支持0~4。</td>
105 <td>INT64</td>105 <td>INT64</td>
106 <td>-</td>106 <td>-</td>
107 <td>-</td>107 <td>-</td>
@@ -170,7 +170,7 @@ aclnnStatus aclnnSphericalHarmonicsBwd(
170 <tr>170 <tr>
171 <td>ACLNN_ERR_PARAM_INVALID</td>171 <td>ACLNN_ERR_PARAM_INVALID</td>
172 <td>161002</td>172 <td>161002</td>
173- <td>dirs、coeffs、vColors、vDirs、vCoeffs的数据类型和数据格式不在支持的范围内，或degree不在0~3范围内。</td>173+ <td>dirs、coeffs、vColors、vDirs、vCoeffs的数据类型和数据格式不在支持的范围内，或degree不在0~4范围内。</td>
174 </tr>174 </tr>
175 </tbody>175 </tbody>
176 </table>176 </table>