new default UV coordinates on some mesh primitives are poor/missing #47478

New Issue

Francis OBrien · 2016-02-18T18:02:42+01:00

Francis OBrien commented

2016-02-18 18:02:42 +01:00

System Information
Windows 10, Nvidia Geforce 860m, 16gig of ram

Blender Version
2.76.11

Short description of error

The new default UV coordinates on some mesh primitives are poor/missing.

The uv sphere and ico sphere have messy uvs. I think the range should also be limited to 0 to 1.
I've attempted to attached pics.

Torus and Suzanne are missing uvs altogether.

I could mock better uvs if that helps?

cheers

**System Information** Windows 10, Nvidia Geforce 860m, 16gig of ram **Blender Version** 2.76.11 **Short description of error** The new default UV coordinates on some mesh primitives are poor/missing. The uv sphere and ico sphere have messy uvs. I think the range should also be limited to 0 to 1. I've attempted to attached pics. Torus and Suzanne are missing uvs altogether. I could mock better uvs if that helps? cheers ![blender_uv_ico_sphere.jpg](https://archive.blender.org/developer/F283752/blender_uv_ico_sphere.jpg) ![blender_uv_sphere.jpg](https://archive.blender.org/developer/F283754/blender_uv_sphere.jpg)

Francis OBrien commented

2016-02-18 18:02:42 +01:00

Changed status to: 'Open'

Francis OBrien commented

2016-02-18 18:02:42 +01:00

Added subscriber: @Francis-4

Bastien Montagne commented

2016-02-19 12:12:26 +01:00

Added subscriber: @mont29

Bastien Montagne commented

2016-02-19 12:12:26 +01:00

Thanks for the report, but this is no bug really, more a set of TODOs. There are several issues here:

UVs of spheres have to be generated procedurally, since user can choose the number of subdivisions etc. If you check results of some other unwrapping methods in Blender, none is really better for spheres here - and quite a few also go outside of normalized coordinates. Not to say this is good result, rather than it’s a TODO to enhance our procedural spherical unwrapping.
Monkey we'd need a hard-coded UVmap. Also a nice TODO.
Torus is another problem, it’s actually generated from python, so ,it cannot really use same system as other primitives here, again a nice TODO to add its own default UVs of course.

The last two points seems to be good candidates for our 'quick hacks', will add those. First point is most likely more involved (especially for icosphere).

Thanks for the report, but this is no bug really, more a set of TODOs. There are several issues here: * UVs of spheres have to be generated procedurally, since user can choose the number of subdivisions etc. If you check results of some other unwrapping methods in Blender, none is really better for spheres here - and quite a few also go outside of normalized coordinates. Not to say this is good result, rather than it’s a TODO to enhance our procedural spherical unwrapping. * Monkey we'd need a hard-coded UVmap. Also a nice TODO. * Torus is another problem, it’s actually generated from python, so ,it cannot really use same system as other primitives here, again a nice TODO to add its own default UVs of course. The last two points seems to be good candidates for our 'quick hacks', will add those. First point is most likely more involved (especially for icosphere).

Bastien Montagne commented

2016-02-19 12:20:03 +01:00

Added #47488 and #47489 quick hacks.

Francis OBrien commented

2016-02-19 12:33:05 +01:00

Hi Bastien

Yes users complicate things! And I can see it's more involved from a coding point of view when it comes to changing the parameters on primitives.

I'll keep a look out for these todos as it's something I've been looking forward to in Blender for over 10 years.

cheers

Hi Bastien Yes users complicate things! And I can see it's more involved from a coding point of view when it comes to changing the parameters on primitives. I'll keep a look out for these todos as it's something I've been looking forward to in Blender for over 10 years. cheers

peter lu commented

2016-08-27 00:40:46 +02:00

Added subscriber: @pdlla

peter lu self-assigned this 2016-08-27 00:40:46 +02:00

peter lu commented

2016-08-27 00:40:46 +02:00

I'm giving this task a go together with the #47488

Chau commented

2016-08-27 01:40:45 +02:00

Added subscriber: @ChauHo

Chau commented

2016-08-27 01:40:45 +02:00

For UV sphere, I think should use this coordinate pattern:

For UV sphere, I think should use this coordinate pattern: ![uvSphereCoord.png](https://archive.blender.org/developer/F344142/uvSphereCoord.png)

Chau commented

2016-08-27 04:28:35 +02:00

For ico sphere I think can use this coordinate system:

Subdivide triangles for other sub division level

For ico sphere I think can use this coordinate system: ![icoSphereCoord.png](https://archive.blender.org/developer/F344188/icoSphereCoord.png) Subdivide triangles for other sub division level

Carlo Andreacchio commented

2016-08-27 06:30:18 +02:00

Added subscriber: @candreacchio

Carlo Andreacchio commented

2016-08-27 06:30:18 +02:00

For UV sphere I would suggest equirectangular mapping

Chau commented

2016-08-29 14:17:47 +02:00

Carlo, like this map:

Carlo, like this map: ![uvSphereCoordRect.png](https://archive.blender.org/developer/F347037/uvSphereCoordRect.png)

peter lu commented

2016-09-01 18:19:03 +02:00

Ok, about to submit my first patch! I went for the map that you suggested Chau. I hardcoded the UV map so it's dependent on the face/loop order. I hope that's OK.

The code below generates a map close to that map. I hard coded its output, and modified it to be exact. The function is totally useable but the exact map is better. Including this here here in case I ever need it again (rather than committing it).



/* unused function for assisting in generating hardcoded icosphere UVs 
 * note, using this function in BM_mesh_calc_uvs_sphere will give acceptable results for icosphere
 */
static void bm_mesh_calc_uvs_sphere_ico_face(BMFace *f, const int cd_loop_uv_offset)
{
    float *uvs[4];
    BMLoop *l;
    BMIter iter;
    float dx;
    int loop_index, loop_index_max_x;
    
    BLI_assert(f->len <= 4);
    
    /* check if face has vertex at pole, if so compute a nearbye point */
    int zcount = 0;
    float avgx = 0, avgy = 0;
    BM_ITER_ELEM_INDEX (l, &iter, f, BM_LOOPS_OF_FACE, loop_index) {
        if(l->v->co[0] == 0 && l->v->co[1] == 0){
            zcount++;
        } else {
            avgx += l->v->co[0];
            avgy += l->v->co[1];
        }
    }
    avgx /= (f->len - zcount);
    avgy /= (f->len - zcount);
    
    BM_ITER_ELEM_INDEX (l, &iter, f, BM_LOOPS_OF_FACE, loop_index) {
        MLoopUV *luv = BM_ELEM_CD_GET_VOID_P(l, cd_loop_uv_offset);
        float x = l->v->co[0];
        float y = l->v->co[1];
        float z = l->v->co[2];
        
        float len = sqrtf(x * x + y * y + z * z);
        float theta;
        /* use neigboring point to compute angle for poles */
        if(x == 0 && y == 0){
            theta = atan2f(avgx,avgy);
        }else{
            theta = atan2f(x,y);
        }
        float phi = saacos(z / len);
        luv->uv[0] = 0.5f + theta / (float)M_PI / 2.0f;
        luv->uv[1] = 1.0f -  phi / (float)M_PI;
        
        uvs[loop_index] = luv->uv;
    }
    
    
    /* Fix awkwardly-wrapping UVs */
    loop_index_max_x = 0;
    for (loop_index = 1; loop_index < f->len; loop_index++) {
        if (uvs[loop_index][0] > uvs[loop_index_max_x][0]) {
            loop_index_max_x = loop_index;
        }
    }
    for (loop_index = 0; loop_index < f->len; loop_index++) {
        if (loop_index != loop_index_max_x) {
            dx = uvs[loop_index_max_x][0] - uvs[loop_index][0];
            if (dx > 0.5f) {
                uvs[loop_index][0] += 1;
            }
        }
    }
    
    /* scale the UVs to be equilateral */
    for (loop_index = 0; loop_index < f->len; loop_index++) {
        uvs[loop_index][0] = uvs[loop_index][0]*10.0f/11.0f-0.05f; /* this is slightly off */
        uvs[loop_index][1] *= 0.47238f;
    }
}

![Screen Shot 2016-09-01 at 9.11.48 AM.png](https://archive.blender.org/developer/F349803/Screen_Shot_2016-09-01_at_9.11.48_AM.png) Ok, about to submit my first patch! I went for the map that you suggested Chau. I hardcoded the UV map so it's dependent on the face/loop order. I hope that's OK. The code below generates a map close to that map. I hard coded its output, and modified it to be exact. The function is totally useable but the exact map is better. Including this here here in case I ever need it again (rather than committing it). ``` /* unused function for assisting in generating hardcoded icosphere UVs * note, using this function in BM_mesh_calc_uvs_sphere will give acceptable results for icosphere */ static void bm_mesh_calc_uvs_sphere_ico_face(BMFace *f, const int cd_loop_uv_offset) { float *uvs[4]; BMLoop *l; BMIter iter; float dx; int loop_index, loop_index_max_x; BLI_assert(f->len <= 4); /* check if face has vertex at pole, if so compute a nearbye point */ int zcount = 0; float avgx = 0, avgy = 0; BM_ITER_ELEM_INDEX (l, &iter, f, BM_LOOPS_OF_FACE, loop_index) { if(l->v->co[0] == 0 && l->v->co[1] == 0){ zcount++; } else { avgx += l->v->co[0]; avgy += l->v->co[1]; } } avgx /= (f->len - zcount); avgy /= (f->len - zcount); BM_ITER_ELEM_INDEX (l, &iter, f, BM_LOOPS_OF_FACE, loop_index) { MLoopUV *luv = BM_ELEM_CD_GET_VOID_P(l, cd_loop_uv_offset); float x = l->v->co[0]; float y = l->v->co[1]; float z = l->v->co[2]; float len = sqrtf(x * x + y * y + z * z); float theta; /* use neigboring point to compute angle for poles */ if(x == 0 && y == 0){ theta = atan2f(avgx,avgy); }else{ theta = atan2f(x,y); } float phi = saacos(z / len); luv->uv[0] = 0.5f + theta / (float)M_PI / 2.0f; luv->uv[1] = 1.0f - phi / (float)M_PI; uvs[loop_index] = luv->uv; } /* Fix awkwardly-wrapping UVs */ loop_index_max_x = 0; for (loop_index = 1; loop_index < f->len; loop_index++) { if (uvs[loop_index][0] > uvs[loop_index_max_x][0]) { loop_index_max_x = loop_index; } } for (loop_index = 0; loop_index < f->len; loop_index++) { if (loop_index != loop_index_max_x) { dx = uvs[loop_index_max_x][0] - uvs[loop_index][0]; if (dx > 0.5f) { uvs[loop_index][0] += 1; } } } /* scale the UVs to be equilateral */ for (loop_index = 0; loop_index < f->len; loop_index++) { uvs[loop_index][0] = uvs[loop_index][0]*10.0f/11.0f-0.05f; /* this is slightly off */ uvs[loop_index][1] *= 0.47238f; } } ```

peter lu commented

2016-09-01 18:37:10 +02:00

This is the modified BM_mesh_calc_uvs_sphere function for outputting icosphere UVs to console. Again, putting it here instead of committing.


void BM_mesh_calc_uvs_sphere(BMesh *bm, const short oflag)
{
	BMFace *f;
	BMIter iter, iter2;

	const int cd_loop_uv_offset = CustomData_get_offset(&bm->ldata, CD_MLOOPUV);

	BLI_assert(cd_loop_uv_offset != -1); /* caller is responsible for giving us UVs */

	BM_ITER_MESH (f, &iter, bm, BM_FACES_OF_MESH) {
		if (!BMO_face_flag_test(bm, f, oflag))
			continue;

		bm_mesh_calc_uvs_sphere_face(f, cd_loop_uv_offset);
	}
    
    /* print UVs for icosphere
    float icouvs[20][3][2];
    int find = 0;
    BM_ITER_MESH (f, &iter, bm, BM_FACES_OF_MESH) {
        if (!BMO_face_flag_test(bm, f, oflag))
            continue;
        
        BMLoop *l;
        int loop_index;
        int vind = 0;
        BM_ITER_ELEM_INDEX (l, &iter2, f, BM_LOOPS_OF_FACE, loop_index) {
            MLoopUV *luv = BM_ELEM_CD_GET_VOID_P(l, cd_loop_uv_offset);
            icouvs[find][vind][0] = luv->uv[0];
            icouvs[find][vind][1] = luv->uv[1];
            vind++;
        }
        find++;
    }
    for(int i = 0; i < 20; i++)
    {
        printf("{{%ff,%ff},{%ff,%ff},{%ff,%ff}},\n",icouvs[i][0][0],icouvs[i][0][1],icouvs[i][1][0],icouvs[i][1][1],icouvs[i][2][0],icouvs[i][2][1]);
    }*/
}

This is the modified BM_mesh_calc_uvs_sphere function for outputting icosphere UVs to console. Again, putting it here instead of committing. ``` void BM_mesh_calc_uvs_sphere(BMesh *bm, const short oflag) { BMFace *f; BMIter iter, iter2; const int cd_loop_uv_offset = CustomData_get_offset(&bm->ldata, CD_MLOOPUV); BLI_assert(cd_loop_uv_offset != -1); /* caller is responsible for giving us UVs */ BM_ITER_MESH (f, &iter, bm, BM_FACES_OF_MESH) { if (!BMO_face_flag_test(bm, f, oflag)) continue; bm_mesh_calc_uvs_sphere_face(f, cd_loop_uv_offset); } /* print UVs for icosphere float icouvs[20][3][2]; int find = 0; BM_ITER_MESH (f, &iter, bm, BM_FACES_OF_MESH) { if (!BMO_face_flag_test(bm, f, oflag)) continue; BMLoop *l; int loop_index; int vind = 0; BM_ITER_ELEM_INDEX (l, &iter2, f, BM_LOOPS_OF_FACE, loop_index) { MLoopUV *luv = BM_ELEM_CD_GET_VOID_P(l, cd_loop_uv_offset); icouvs[find][vind][0] = luv->uv[0]; icouvs[find][vind][1] = luv->uv[1]; vind++; } find++; } for(int i = 0; i < 20; i++) { printf("{{%ff,%ff},{%ff,%ff},{%ff,%ff}},\n",icouvs[i][0][0],icouvs[i][0][1],icouvs[i][1][0],icouvs[i][1][1],icouvs[i][2][0],icouvs[i][2][1]); }*/ } ```

peter lu commented

2016-09-07 20:02:21 +02:00

Added UV sphere map using equirectangular map. I would have liked to use quads on the top and bottom row but those faces only have 3 vertices so I went with the spiky triangles.

The rounding is done by scaling the x coordinate by an arbitrary factor of sin(phi*0.7) i.e. the code looks something like

        theta = atan2f(y,x);
        float phi = acos(z / len);
        float phiScaled = (phi-M_PI/2.0f)*0.7f + M_PI/2.0f;
        luv->uv[0] = 0.5f + sin(phiScaled) * theta / (float)M_PI / 2.0f;
        luv->uv[1] = 1.0f -  phiScaled / (float)M_PI;

Added UV sphere map using equirectangular map. I would have liked to use quads on the top and bottom row but those faces only have 3 vertices so I went with the spiky triangles. ![Screen Shot 2016-09-07 at 10.54.51 AM.png](https://archive.blender.org/developer/F355553/Screen_Shot_2016-09-07_at_10.54.51_AM.png) The rounding is done by scaling the x coordinate by an arbitrary factor of sin(phi*0.7) i.e. the code looks something like ``` theta = atan2f(y,x); float phi = acos(z / len); float phiScaled = (phi-M_PI/2.0f)*0.7f + M_PI/2.0f; luv->uv[0] = 0.5f + sin(phiScaled) * theta / (float)M_PI / 2.0f; luv->uv[1] = 1.0f - phiScaled / (float)M_PI; ```

peter lu commented

2016-09-07 20:50:10 +02:00

Adding UVs to Monkey and Torus are #47488 and #47489 respectively. D2202 and D2186 add better UVs for icosphere and uvsphere respectively. My first 2 patches :D! Can I/someone mark this task as resolved if/when D2202 and D2186 are accepted?

Adding UVs to Monkey and Torus are #47488 and #47489 respectively. [D2202](https://archive.blender.org/developer/D2202) and [D2186](https://archive.blender.org/developer/D2186) add better UVs for icosphere and uvsphere respectively. My first 2 patches :D! Can I/someone mark this task as resolved if/when [D2202](https://archive.blender.org/developer/D2202) and [D2186](https://archive.blender.org/developer/D2186) are accepted?

Francis OBrien commented

2016-09-08 13:16:42 +02:00

@ peter lu

looking good so far. I've never mapped primitives using your solutions but they look interesting and useful.

However can you keep the uv coords between 0 and 1? I think it's a better solution.

cheers

@ peter lu looking good so far. I've never mapped primitives using your solutions but they look interesting and useful. However can you keep the uv coords between 0 and 1? I think it's a better solution. cheers

Francis OBrien commented

2016-09-08 13:19:02 +02:00

here are a few default mesh and uv examples from Maya. I'm not saying copy them :) But this is how the competition does it

here are a few default mesh and uv examples from Maya. I'm not saying copy them :) But this is how the competition does it ![sphere_pinched_at_poles_uvs.jpg](https://archive.blender.org/developer/F356076/sphere_pinched_at_poles_uvs.jpg) ![sphere_sawtooth_at_poles_uvs.jpg](https://archive.blender.org/developer/F356078/sphere_sawtooth_at_poles_uvs.jpg) ![torus_uvs.jpg](https://archive.blender.org/developer/F356080/torus_uvs.jpg)

peter lu commented

2016-09-08 17:48:01 +02:00

@Francis-4

It shifts the uvs to fit in [0,1] now :), I just used an old screenshot (TBH though I was almost about to be too lazy to do this XD).

Thanks for those reference images. Those are actually special cases of my current implementation. Maybe I'll add them as input parameters to the UV spehere.

@Francis-4 It shifts the uvs to fit in [0,1] now :), I just used an old screenshot (TBH though I was almost about to be too lazy to do this XD). Thanks for those reference images. Those are actually special cases of my current implementation. Maybe I'll add them as input parameters to the UV spehere.

Chau commented

2016-09-09 09:39:34 +02:00

Two bottom pictures from Maya follow my first maps for UV sphere and torus.

blender-admin commented

2016-09-24 16:08:10 +02:00

This issue was referenced by a070a5befa

This issue was referenced by a070a5befa1110120547333d84c6e67cae53d648

Brecht Van Lommel commented

2016-09-24 16:40:44 +02:00

Changed status from 'Open' to: 'Resolved'

Brecht Van Lommel closed this issue

2016-09-24 16:40:44 +02:00

Sign in to join this conversation.

No Label

Download

What's New

Blender Studio

Manual

Developers Blog

Documentation

Benchmark

Blender Conference

Development Fund

One-time Donations

new default UV coordinates on some mesh primitives are poor/missing #47478